1 2 3 4 1 Magmatic evolution of the Campi Flegrei and 5 6 7 2 Volcanic Fields, , based on interpretation of data 8 9 3 from well-constrained melt inclusions 10 11 4 Rosario Esposito1, Kimberly Badescu1, Matthew Steele-MacInnis2, Claudia Cannatelli3, 12 13 5 Benedetto De Vivo4, Annamaria Lima5, Robert J. Bodnar6, Craig E. Manning1 14 15 6 1Department of Earth, Planetary and Space Science, UCLA, Los Angeles, CA 16 7 90095-1567 USA email: [email protected] 17 2 18 8 Department of Earth and Atmospheric Sciences, University of Alberta, 19 9 Edmonton, Alberta Canada T6G 2E3 20 10 3Department of Geology and Andean Geothermal Center of Excellence (CEGA), 21 11 Universidad de Chile, Santiago, Chile 22 23 12 Università telematica Pegaso, Piazza Trieste e Trento 48, 80132 Napoli & 24 13 Benecon Scarl, Via S. Maria di Costantinopoli 104, 80138, Napoli 25 14 5Dipartimento di Scienze della Terra, delle Risorse e dell’Ambiente, Università di 26 27 15 Napoli Federico II, Via Cintia snc, 80126 Napoli, Italy 28 16 6Fluids Research Laboratory, Department of Geosciences, Virginia Tech, 29 17 Blacksburg, VA 24061 USA 30 18 31 32 33 34 19 Abstract 35 20 One of the main goals of studying melt inclusions (MI) is to constrain the pre- 36 37 21 eruptive physical and chemical processes that have occurred in a magma reservoir at the 38 39 22 micro-scale. Recently, several studies that focused on magmatic differentiation of 40 23 volcanic systems produced detailed interpretations based on data from MI trapped at 41 42 24 different times and locations in the plumbing system. Ideally, MI data should be collected 43 25 and tested following the melt inclusion assemblage (MIA) protocol that consists of 44 45 26 studying and analyzing groups of MI that were trapped at the same time, and, thus, at the 46 47 27 same chemical and physical conditions. However, the rarity of MIA in juvenile volcanic 48 28 phenocrysts precludes this methodology in many cases, leading to uncertainty concerning 49 50 29 the validity of the MI as recorders of pre-eruptive conditions. 51 52 53 54 1 55 56 57 58 59 30 In this study, we focused on MI from the Campi Flegrei (CF) and Procida Island 60 61 31 (PI) volcanic systems in , including data from this study and data from the 62 63 32 literature. The database included MI hosted in sanidine, clinopyroxene, plagioclase, 64 33 biotite and olivine, and, thus, represents melts trapped at various stages in the overall 65 66 34 differentiation process. We developed a protocol to select the most reliable MI from a 67 35 dataset associated with a single magmatic system. As a first step we compare MI data 68 69 36 with bulk rock data for the same magmatic system. This comparison reveals that most MI 70 71 37 show major element compositions that fall within or close to the range for bulk rocks – 72 38 these MI are considered to be “normal”. Some MI show anomalous compositions and are 73 74 39 not representative of the melt in equilibrium with the phenocryst host and were excluded 75 40 from the data set. In the second step we selected only bubble-free MI from the previously 76 77 41 identified “normal” MI to interpret the volatile evolution. In the third step we compare 78 42 compositions of the “normal” bubble-free MI to compositions predicted by rhyolite- 79 80 43 MELTS simulations, assuming a variety of initial conditions. 81 82 44 Comparison of data obtained from basaltic-trachybasaltic MI with rhyolite- 83 45 MELTS predictions indicates that one group of MI records the geochemical evolution of 84 85 46 a volatile-saturated magma differentiating by polybaric fractional crystallization from 86 47 ≥200 MPa (≥7.5 km) to 30 MPa (~1 km). Another group of MI records recharge of the 87 88 48 magma chamber by a primitive basaltic magma that mixes with the preexisting primitive 89 90 49 trachybasaltic magma before eruption. Extensive isobaric crystallization of the 91 50 trachybasaltic magmas at ~ 7.5 km beneath CF may have generated trachytic-phonolitic 92 93 51 magmas, such as those associated with the Neapolitan Yellow (NYT) that is 94 52 characterized by a relatively high H O content. These volatile-saturated trachytic- 95 2 96 53 phonolitic magmas likely trigger high-magnitude eruptions during their ascent to the 97 98 54 surface. 99 55 100 101 56 Keywords: melt inclusion, volatiles, Phlegrean Fields, degassing path, rhyolite-MELTS 102 103 104 105 106 107 108 109 110 2 111 112 113 114 115 116 57 Introduction 117 118 58 Campi Flegrei (CF) and the Island of Procida (IP) are well known recently active 119 59 volcanic complexes in the Campanian Plain of southern Italy that represent a volcanic 120 121 60 hazard owing to their proximity to and the towns surrounding Bay (~1.5 122 61 million inhabitants). The area has been the site of volcanic activity for more than 60 ka 123 124 62 (De Vivo et al., 2010; Orsi et al., 1999; Pappalardo et al., 1999). The most catastrophic 125 63 eruption at CF is the Campanian Ignimbrite (e.g., Orsi et al., 1996), generating 300 km3 126 127 64 dense rock equivalent [DRE; (Fedele et al., 2007)], the source of which remains under 128 129 65 debate (De Vivo et al., 2010; De Vivo et al., 2001; Rolandi et al., 2003). The later 130 66 eruption of the Neapolitan Yellow Tuff (NYT) was extremely explosive and generated 40 131 132 67 km3 DRE of eruptive products (Orsi et al., 1992; Wohletz et al., 1995). The most recent 133 68 eruption occurred in 1538 and formed the cinder cone (Fig. 1). 134 135 69 The CF area has in recent decades attracted the interest of geoscientists owing to 136 137 70 activity that consists of episodic slow ground uplift, followed by more rapid deflation 138 71 (bradyseism). In recent years, the ground in the central part of the CF has been inflating, 139 140 72 highlighting the critical need for understanding whether this activity is a precursor of an 141 73 imminent eruption (e.g., Chiodini et al., 2015). Ground deformation and seismicity at CF 142 143 74 are also associated with intense fumarolic and hydrothermal activity, concentrated mostly 144 145 75 in the crater of at Pozzuoli (Fig. 1). The relatively slow ground uplift may or 146 76 may not represent a precursor of a volcanic eruption, depending on the cause of the 147 148 77 inflation (Bodnar et al., 2007; Lima et al., 2009; Moretti et al., 2017). Though various 149 78 models have been proposed to explain bradyseism at CF, it is now becoming clear that 150 151 79 magmatic fluids released from deep and/or shallow magmas play an important role in 152 153 80 ground movement, and the occurrence of ground movement does not require that a new 154 81 batch of magma is intruding below CF (Lima et al., 2009; Lima et al., 2017). 155 156 82 The composition of exsolved magmatic fluids injected into the overlying 157 83 hydrothermal system is controlled by the chemical evolution of volatiles during magma 158 159 84 evolution (crystallization, mixing, and crustal assimilation). The compositions of MI 160 85 hosted in olivine from CF and IP suggest that trachybasaltic parental magmas are the 161 162 86 source of activity in both areas (Cannatelli et al., 2007; Mangiacapra et al., 2008; 163 164 87 Mormone et al., 2011; Esposito et al., 2011). However, the isotopic compositions of 165 166 3 167 168 169 170 171 88 mafic bulk rocks suggest that the two areas experienced different magma evolution 172 173 89 histories (e.g., Cannatelli, 2012; D'Antonio et al., 2007; Peccerillo, 2017 and references 174 175 90 therein). While major and trace element trends in both localities are dominantly 176 91 controlled by fractional crystallization, the variability of some elements such as 177 178 92 potassium , the isotopic signature of the bulk rocks, and the composition and zoning of 179 93 juvenile mineral phases combined suggest that other open-system magmatic processes 180 181 94 also contributed to observed geochemical trends. The geochemical variability, evolution, 182 183 95 and the importance of open-system processes associated with the generation and the 184 96 differentiation of these magmas continue to be debated (Peccerillo, 2017 and references 185 186 97 therein). 187 98 The pre-eruptive volatile contents of magmas associated with CF and IP have 188 189 99 been studied by many workers by analyzing the glass of melt inclusions (MI) (Arienzo et 190 100 al., 2016; Arienzo et al., 2010; Cannatelli et al., 2007; Cecchetti et al., 2001; Cecchetti et 191 192 101 al., 2002-2003; Cipriani et al., 2008; Esposito et al., 2011; Fourmentraux et al., 2012; 193 194 102 Mangiacapra et al., 2008; Mormone et al., 2011; Roach, 2005; Stock et al., 2016). MI 195 103 provide information that is not easily obtained by studying bulk rocks or active 196 197 104 fumaroles. MI often preserve pre-eruptive volatile contents because host phenocrysts act 198 105 as an insulating capsule and MI are not affected by degassing of the surrounding magma 199 200 106 that likely occurs during ascent and crystallization (Metrich and Wallace, 2008). 201 202 107 However, interpretation of volatile data from MI, and thus the modeling of magma 203 108 dynamics, is based on the assumption that MI record the composition of the melt from 204 205 109 which the host minerals formed. However, MI may not preserve the original melt 206 110 composition if (1) the composition of the glass phase does not represent the composition 207 208 111 of the melt that was originally trapped owing to a variety of processes [e.g., post 209 210 112 entrapment crystallization, diffusion of elements, formation of a vapor bubble, 211 113 decrepitation (Cannatelli et al., 2016 and references therein)], (2) if boundary layer 212 213 114 processes during trapping modified the composition of the melt adjacent to the growing 214 115 crystal, such that the trapped melt does not represent the bulk (far-field) melt (e.g., Baker, 215 216 116 2008), or (3) the melt that was trapped was generated as a result of disequilibrium 217 218 117 processes (e.g., Danyushevsky et al., 2004). It is important to note that while 219 118 concentrations and ratios of elements that are compatible in the host will be modified by 220 221 222 4 223 224 225 226 227 119 post entrapment crystallization (PEC), PEC does not affect ratios of incompatible 228 229 120 elements in the melt (now glass) (Danyushevsky et al., 2002a; Lima et al., 2003). 230 231 121 The most rigorous and dependable approach to confirm that the composition of a 232 122 MI is representative of the melt from which the host crystal grew is to study groups of MI 233 234 123 that were trapped at the same time, defined as a “melt inclusion assemblage” (MIA; 235 124 Bodnar and Student, 2006). All MI within an MIA would have been trapped at the same 236 237 125 temperature and pressure, and all would have trapped a melt of the same composition. As 238 239 126 such all of the MI within the MIA should show consistent phase behavior and chemistry 240 127 (Bodnar and Student, 2006). While studying MIAs can provide unarguable evidence that 241 242 128 the MI have trapped and preserved the original melt from which the host crystal was 243 129 growing, melt inclusion assemblages are uncommon in most samples, and workers have 244 245 130 rarely reported melt inclusion assemblages in studies of MI from CF and IP owing to 246 131 their scarcity (Esposito et al., 2014). Also, the few MIA identified by Esposito et al. 247 248 132 (2014) are comprised of small MI (<20 µm), and therefore challenging or impossible to 249 250 133 analyze by secondary ion mass spectrometry (SIMS) and/or Fourier-transform infrared 251 134 (FTIR) spectroscopy. 252 253 135 As an alternative to the MIA approach to test that compositions of MI represent 254 136 the original melt that was present when the phenocryst was growing, the composition of 255 256 137 MI may be compared to the bulk rock compositions of samples from the same magmatic 257 258 138 system (Danyushevsky et al., 2000). This comparison can help identify MI compositions 259 139 that are anomalous relative to general geochemical trends defined by bulk rocks. While 260 261 140 both the MIA approach and comparison of MI compositions with bulk rock compositions 262 141 can be applied to determine if the volatile contents of MI represent the volatile content of 263 264 142 the original, pre-eruptive melt, we note that most previous studies of MI from CF and IP 265 266 143 have not reported or discussed methods to test whether the major elements and volatile 267 144 contents of MI record the original, unmodified composition. Here, we use the term 268 269 145 “representativeness” to identify a MI that traps and preserves a sample of the melt that 270 146 was in equilibrium with the host phenocryst at the time of trapping. In other words, MI 271 272 147 represent a “snapshot” of the liquid (melt) composition during magmatic differentiation, 273 274 148 representing melt that was in equilibrium with crystals ± vapor. One of the main factors 275 149 that limits our ability to determine whether volatile contents of MI from CF and IP 276 277 278 5 279 280 281 282 283 150 reported in the literature represent the original volatile content of the trapped melt is that 284 285 151 contributions of the vapor bubble in the MI to the total volatile content of the MI are not 286 287 152 discussed or considered, even though vapor bubbles are reported in the descriptions of MI 288 153 petrography. Recent studies confirm that analyzing only the glass phase of a bubble- 289 290 154 bearing MI and ignoring the CO2 content of the bubble significantly underestimates the 291 155 total CO2 (and other volatiles) content of the trapped melt (e.g., Aster et al., 2016; 292 293 156 Esposito et al., 2016; Moore et al., 2015). Because most of the CO2 is contained in the 294 295 157 vapor bubble of a bubble-bearing MI, a more realistic and accurate estimate of the CO2 296 158 content of the trapped melt is obtained if only the bubble, rather than only the glass, is 297 298 159 analyzed. Concentrations of other volatiles, including H2O and S, will similarly be 299 160 underestimated if the volatile content of the bubbles is not considered (Esposito et al., 300 301 161 2016). 302 162 Given these concerns, it is appropriate to re-evaluate MI data from CF and IP 303 304 163 because published reports often do not specify whether or not the MI that were studied 305 306 164 contain vapor bubbles. In addition, MI compositions are often not compared with bulk 307 165 rock compositions and geochemical trends are often not described. Moreover, if MI 308 309 166 compositions do not follow the main bulk rock trends, the significance of these 310 167 anomalous MI compositions is not considered. It is important to note that all, or most, of 311 312 168 the MI in the studies we cited (and in one case, conducted) may be representative of the 313 314 169 bulk melt during differentiation, but the representativeness has not been rigorously 315 170 evaluated. Thus, it remains unclear whether the variations in volatile contents observed in 316 317 171 MI represent variations in the composition of melts in equilibrium with the growing 318 172 crystal at the time of trapping, or instead are due to modifications after trapping. 319 320 173 In this study, we review published data for MI from the CF and IP volcanic areas 321 322 174 (Arienzo et al., 2016; Arienzo et al., 2010; Cannatelli et al., 2007; Cecchetti et al., 2002- 323 175 2003; Cipriani et al., 2008; Esposito et al., 2011; Fourmentraux et al., 2012; Mangiacapra 324 325 176 et al., 2008; Mormone et al., 2011; Roach, 2005; Stock et al., 2016), and we present new 326 177 geochemical data obtained from MI from the CF and IP volcanic fields. The new data are 327 328 178 from representative samples from the Neapolitan Yellow Tuff (NYT), Solchiaro, , 329 330 179 Agnano-Monte Spina (AMS), Solfatara, and Fossa Lupara eruptions. These new data 331 180 supplement the database of MI compositions from the literature. We selected MI from the 332 333 334 6 335 336 337 338 339 181 NYT, Bacoli, and Fossa Lupara eruptions because, to our knowledge, MI from these 340 341 182 eruptions have not yet been studied. Moreover, the selected samples represent a 342 343 183 significant range in time (~ 23.6 ka to 3.7 ka) and space within the CF and IP (Fig. 1). By 344 184 conducting new analyses within a rigorous analytical protocol, our goal was to develop a 345 346 185 better understanding of the reliability of MI as recorders of the pre-eruptive magma 347 186 history at CF and IP. We compared compositions of MI from this study with those of 348 349 187 bulk rocks and MI from the literature to better understand the correlation between MI and 350 351 188 bulk rock compositions (Appendix Table D). The compiled literature data, combined 352 189 with our new data, allow examination of possible correlations between the major element 353 354 190 and volatile contents of MI and the compositions of host phenocrysts (mainly olivine, 355 191 clinopyroxene, and sanidine). In addition, we compared compositions of experimentally 356 357 192 reheated MI with data from naturally quenched MI. The selected MI data for CF and IP 358 193 and mineral data were also compared with differentiation trends and mineral 359 360 194 compositions predicted using rhyolite-MELTS simulations to better understand if 361 362 195 crystallization occurred under volatile-saturated conditions (Ghiorso and Gualda, 2015). 363 196 Finally, we describe in detail a protocol to test the representativeness of MI in the 364 365 197 absence of melt inclusion assemblages. We demonstrate that re-evaluating MI data 366 198 following the method employed here reduces uncertainty and instills confidence that the 367 368 199 MI compositions are representative of the original pre-eruptive melt. 369 370 200 371 372 373 201 Geology and geochronology of Campi Flegrei and Procida 374 375 202 376 Island 377 203 Campi Flegrei (CF) and Procida Island (IP) are part of the Phlegrean Volcanic 378 379 204 District (PVD; Orsi et al., 1996). The PVD, along with the Somma-Vesuvius volcanic 380 381 205 district, is part of the Campanian Comagmatic Province, which in turn is part of the 382 206 circum-Tyrrhenian magmatic event that began in the Plio-Quaternary (Peccerillo, 1999 383 384 207 and references therein). Campanian magmatism is associated with NW-SE and NE-SW 385 208 trending normal faults that reflect the extensional/trans-tensional tectonic regime of the 386 387 209 Campanian margin (Milia, 2010 and references therein). The magmas generated in the 388 389 390 7 391 392 393 394 395 210 CF and IP share a common origin dating back at least 40 ka (De Astis et al., 2004). The 396 397 211 oldest volcanic products in the entire Campanian Plain do not outcrop but, rather, were 398 399 212 collected from drill cores and give ages of 2.0 ± 0.4 Ma (Barbieri et al., 1979), ~ 1 Ma 400 213 (Ippolito et al., 1975) and ~ 0.4 Ma (Brocchini et al., 2001). 401 402 403 214 Campi Flegrei 404 215 The Campi Flegrei (CF) volcanic field is located NW of Naples and is 405 406 216 characterized by the presence of several volcanic cones, rings, and structures. The 407 217 volcanic framework at CF is interpreted by some researchers as representing a nested and 408 409 218 resurgent caldera that is partially submerged under the Bay of Pozzuoli (Orsi et al., 1995; 410 411 219 Orsi et al., 1992; Orsi et al., 1996; Perrotta et al., 2006) (Fig. 1). In this model, the outer 412 220 caldera rim was formed during eruption of the Campanian Ignimbrite (CI), the largest 413 414 221 eruption in the entire Campanian Plain, that occurred at ~39 ka (De Vivo et al., 2001; 415 222 Rolandi et al., 2003), and the inner caldera was formed later by the Neapolitan Yellow 416 417 223 Tuff (NYT) eruption at ~15 ka (Deino et al., 2004). While most workers agree that a 418 419 224 caldera at CF was formed during the NYT eruption, both the existence of an outer caldera 420 225 formed by the CI and the source of the CI are subjects to debate (Scandone et al., 2006). 421 422 226 Rosi and Sbrana (1987), Orsi et al. (1996), and Piochi et al. (2013) suggest that the CI 423 227 was erupted from a vent located in the center of the CF. In contrast, Scandone et al. 424 425 228 (1991) suggest that the CI originated from the depression northeast of Naples. 426 427 229 Another interpretation is that the CI was produced as the result of multiple fissure 428 230 eruptions that occurred along faults associated with the local extensional tectonic regime 429 430 231 (De Vivo et al., 2010; De Vivo et al., 2001; Rolandi et al., 2003). As summarized by De 431 232 Vivo et al. (2010), these eruptions fed by regional faults range in age from > 300 ka to 39 432 433 233 ka. Recent findings based on a new borehole at CF support this latter hypothesis that the 434 435 234 CI originated through fissure vents from outside of the CF (De Natale et al., 2016). In 436 235 particular, these researchers reported that the volume of the NYT is 2.5 larger than that of 437 438 236 the CI, and that the volume of the CI deposits contrasts with the hypothesis that the CI 439 237 formed a caldera structure larger than that associated with the NYT. In addition, we note 440 441 238 that there is some debate concerning the location of both the CI and the NYT caldera rims 442 239 (e.g., Orsi et al., 1996; Perrotta et al., 2006; Rosi and Sbrana, 1987; Vitale and Isaia, 443 444 240 2014). Regardless of the origin of the CI, we do not include data from MI from CI in this 445 446 8 447 448 449 450 451 241 assessment because (1) no basaltic or trachybasaltic MI are reported from this eruption 452 453 242 (Marianelli et al., 2006; Signorelli et al., 1999; Webster et al., 2003) (2) it is generally not 454 455 243 possible to determine from the published studies which MI are bubble-bearing and which 456 244 are not, and (3) the origin of the CI and its possible genetic relationship to CF and PI are 457 458 245 still under debate for the reasons mentioned above. 459 246 The oldest outcropping volcanic products in the CF volcanic field, dated at 58±3 460 461 247 ka, are found only in escarpments in the northern part of CF (Pappalardo et al., 1999). 462 463 248 Activity occurring in the interval between these oldest recognized eruptions and the 464 249 eruption of the CI (~39 ka; De Vivo et al., 2001; Rolandi et al., 2003) is referred to as 465 466 250 pre-CI activity. Pre-CI activity includes the Punta di Marmolite and Cuma lava domes 467 251 (Gillot et al., 1982), two of the few lava domes at CF. Other mostly alkali-trachytic 468 469 252 explosive eruptions have been recognized during the pre-CI activity (Pappalardo et al., 470 253 1999). A lithic breccia associated with an ignimbrite deposit and dated at ~39 ka (De 471 472 254 Vivo et al., 2001; Rolandi et al., 2003) formed after this period of relatively minor 473 474 255 volcanic activity. This lithic breccia, known as the Breccia Museo-Piperno formation, 475 256 outcrops at many localities around the CF and some authors suggest that it represents the 476 477 257 proximal phase of the CI (Orsi et al., 1996; Pappalardo et al., 1999; Perrotta et al., 2006). 478 258 Various estimates of the volume of material erupted during the CI range from 80 km3 479 480 259 (Thunell et al., 1979), to 150 km3 (Civetta et al., 1997), to 200 km3 (Rolandi et al., 2003), 481 3 482 260 to 300 km (Fedele et al., 2007) DRE. The estimated volume is related to the ignimbrite 483 261 dated at ~39 ka by De Vivo et al. (2001). 484 485 262 Following the CI eruption, volcanic activity is characterized by several eruptions 486 263 referred to as “tufi biancastri” (whitish tuffs) that occur mainly in eastern CF (Di 487 488 264 Girolamo et al., 1984). These eruptions span a period from <39 ka to 15 ka, at which time 489 490 265 the NYT erupted. The age of the NYT is controversial because two different analytical 491 266 methods give significantly different ages. The NYT has been dated at 12 ka using the 14C 492 493 267 method (Alessio et al., 1971) and at ~15 ka using the 39Ar/40Ar method (Deino et al., 494 268 2004). As mentioned above, the NYT is interpreted to be the eruption that formed the CF 495 496 269 caldera, and the NYT deposits cover an area of 1000 km2 and consist of 40 km3 DRE 497 498 270 (Orsi et al., 1992; Wohletz et al., 1995). Volcanic activity at CF that occurred after the 499 271 NYT eruption is referred to as “recent activity” by Di Girolamo et al. (1984). Di Vito et 500 501 502 9 503 504 505 506 507 272 al. (1999) grouped post-NYT activity into three epochs of volcanic activity: epoch I 508 509 273 between ~15 and 10.6 ka; epoch II between ~9.6 and 9.1 ka; and epoch III between ~5.5 510 511 274 and 3.5 ka, based on a recalculated age reported by Smith et al. (2011). Di Vito et al. 512 275 (1999) recognized 64 volcanic units that were deposited after the eruption of the NYT. 513 514 276 The common characteristic of this recent activity is that eruptions were mostly explosive, 515 277 generated from monogenetic centers, and produced a small amount of material compared 516 517 278 to the NYT eruption. The eruptions of the first two epochs are located mainly near the 518 519 279 periphery of CF, while more recent eruptions occurred mostly in the interior of CF. 520 280 Recently, Lirer et al. (2011) recognized a new volcanic unit that they named Torretta. 14C 521 522 281 dating of samples from the Torretta unit revealed an age of 2.8 ka (Lirer et al., 2011). The 523 282 Monte Nuovo eruption in 1538 AD represents the only historical eruption at CF, and a 524 525 283 historical chronicle describes in detail the week-long eruption that formed a scoria cone 526 284 (Da Toledo, 1539). 527 528 529 285 Procida 530 531 286 The Island of Procida (IP) represents the products of volcanic activity from five 532 287 monogenetic volcanoes: , Terra Murata, Pozzo Vecchio, Fiumicello, and Solchiaro 533 534 288 (De Astis et al., 2004; Di Girolamo et al., 1984; Di Girolamo and Stanzione, 1973; 535 289 Pescatore and Rolandi, 1981; Rosi et al., 1988a; Rosi et al., 1988b). These volcanoes are 536 537 290 aligned along a NE-SW trending that extends from the Island of 538 539 291 southwest of IP to to the northeast (Fig. 1). The importance of the NE- 540 292 SW transverse faults that are associated with the Appennine (NW-SE) normal fault in the 541 542 293 formation of the Procida Island volcanoes is noted in several studies (Acocella et al., 543 294 1999; De Astis et al., 2004; Orsi et al., 1996). The IP volcanism has likely been active 544 545 295 over the last 80 ka, but no historical volcanic activity has been recorded (De Astis et al., 546 547 296 2004 and references therein). Volcanic deposits resulting from the activity of the five 548 297 volcanoes are interlayered with deposits from higher magnitude eruptions sourced from 549 550 298 Ischia Island and CF. The most recent eruptive materials on Procida are associated with 551 299 the Solchiaro eruption. Paleosols at the base of the Solchiaro deposits have been dated at 552 553 300 17.3 ka and 19.6 ka, respectively by Lirer et al. (1991) and by Alessio et al. (1989). All of 554 301 the paleosol ages were based on radiocarbon (14C) measurements. Also, Alessio et al. 555 556 302 (1989) reported an age of 14.3 ka for a paleosol above the Solchiaro deposits. This 557 558 10 559 560 561 562 563 303 suggests that the last volcanic activity at Procida occurred between 19.6 ka and 14.3 ka. 564 565 304 More recently, Morabito et al. (2014) recalibrated the 19.6 ka age reported by Alessio et 566 567 305 al. (1989) to obtain an age of 23.624 ± 0.33 cal ka BP, moving the age of the Solchiaro 568 306 eruption to an earlier time. 569 570 571 307 Genesis and differentiation of magma at Campi Flegrei and Procida Island 572 308 The CF and IP magmatism is part of the Quaternary magmatism that is 573 574 309 characterized by high K2O contents in relatively primitive melts (Peccerillo, 2017). The 575 310 trend is defined by the potassic series starting from least evolved trachybasalts, 576 577 311 continuing through shoshinites and latites, and ending with trachytes and phonolites 578 579 312 (Suppl. Fig. 1a). There is no correlation between the age and compositional evolution of 580 313 rocks erupted in the Phlegrean area. The more primitive magmas of the last 23.6 ka 581 582 314 erupted along regional normal faults oriented NE-SW, along the NYT caldera border and 583 315 continuing through Procida and Ischia Islands (Acocella and Funiciello, 1999; De Astis et 584 585 316 al., 2004; Orsi et al., 1996). The volume of trachyte is much larger than the combined 586 587 317 volumes of trachybasalt, shoshonite and latite. In particular, the intermediate composition 588 318 rocks represent the smallest volume of erupted products. 589 590 319 Bulk rock major element compositions show trends mostly consistent with simple 591 320 fractional crystallization (D'Antonio, 2011 and references therein), in which olivine is the 592 593 321 predominant mineral phase in the least evolved members. Clinopyroxene is present 594 595 322 throughout the entire differentiation trend and predominates in the intermediate members, 596 323 and sanidine is more abundant in the more evolved stages. Plagioclase also crystallized 597 598 324 along the trend, with accessory phases chromite, Ti-magnetite, apatite and biotite 599 325 observed. Melluso et al. (2012) calculated that ~90% of fractional crystallization (alkali- 600 601 326 feldspar, plagioclase, clinopyroxene and olivine in order of abundance) is required to 602 603 327 generate the trachyte-phonolites from the latitic bulk composition of CF. However, 604 328 simple fractional crystallization models cannot explain either the variability of K for the 605 606 329 least evolved rocks, nor the K-enrichment of some of the more evolved rocks. Other 607 330 elements (e.g., Cl, Sb) show similar anomalous variability and enrichment (Villemant, 608 609 331 1988). The combination of isotopic and trace element data suggests that a mechanism of 610 332 percolation of fluids from the wall rock into the magma reservoir caused selective 611 612 613 614 11 615 616 617 618 619 333 enrichment of K-Sb-Cl during fractional crystallization (Civetta et al., 1991; Villemant, 620 621 334 1988). 622 623 335 In several studies, the large variation of CO2 and H2O concentrations of MI at CF 624 336 and IP is interpreted as the combination of continued degassing of volatile-saturated 625 626 337 magmas during ascent and multiple CO2 fluxing events (Arienzo et al., 2016; Arienzo et 627 338 al., 2010; Mangiacapra et al., 2008; Mormone et al., 2011). In addition, Moretti et al. 628 629 339 (2013) studied major elements and volatiles in trachybasaltic and shoshonitic MI from 630 631 340 Ischia Island (West/South-West from CF and IP) and discussed the importance of CO2 632 341 fluxing for the geochemical evolution of magmas associated with Neapolitan volcanism. 633 634 342 It is important to note that studies advocating the importance of CO2 fluxing corroborate 635 343 the hypothesis of selective enrichment of K, Sb, and Cl by percolation of fluids through 636 637 344 magmas at CF (Civetta et al., 1991; Villemant, 1988). However, D'Antonio and Di 638 345 Girolamo (1994) noted that the selective enrichment of K-Sb-Cl in mafic rocks from IP is 639 640 346 accompanied by the enrichment of other trace elements that are not compatible with 641 642 347 fluids (e.g., Nb). In addition, “shrinkage” bubbles in MI are reported in the MI from 643 348 studies advocating CO2 fluxing, raising concerns about the reliability of the volatile data 644 645 349 recorded by MI (i.e., did the MI trap some CO2 along with melt?). Alternatively, 646 350 D'Antonio and Di Girolamo (1994) invoked mixing between magma batches showing 647 648 351 different degrees of evolution (e.g., trachybasalt and latite) and having different Sr- 649 650 352 isotopic signatures to explain the observed trends. 651 353 There is a general agreement regarding the depth at which extensive 652 653 354 crystallization and possible CO2 fluxing occurred (and is occurring) beneath the CF and 654 355 PI volcanic fields. Seismic reflection data suggest a partial melting zone ~1 km thick at 655 656 356 7.5 km beneath CF. In agreement with geophysical data, data from MI suggest that 657 658 357 extensive crystallization and CO2 fluxing occurred at ~200 MPa (Arienzo et al., 2010; 659 358 Mangiacapra et al., 2008; Moretti et al., 2013). 660 661 359 Several hypotheses have been proposed for magma generation in the Phlegrean 662 360 area. Many researchers argue that magmatism originated from an intraplate-type mantle 663 664 361 source that was subsequently metasomatized by slab-derived fluids (Peccerillo, 2017 and 665 666 362 references therein). Trace element data for CF and IP mafic magmas are enriched in 667 363 LILE and LREE relative to MORB, and the trace element patterns are similar to those of 668 669 670 12 671 672 673 674 675 364 intraplate basalts. In contrast, Ti, Y, and Yb are depleted relative to MORB and the HFSE 676 677 365 show lower degrees of enrichment relative to LILE and LREE, characteristics of an arc- 678 679 366 type signature (De Astis et al., 2004; Peccerillo, 1999 and references therein). 680 367 Volcanic rocks erupted from the CF volcanic field during the last 60 ka show 681 87 86 682 368 highly variable isotopic compositions. In particular, the Sr/ Sr ratio varies from 683 369 ~0.7065 to 0.7086, 143Nd/144Nd varies from ~0.5124 to 0.5128, 206Pb/204Pb from ~18.90 684 685 370 to 19.25, 207Pb/204Pb from ~15.65 to 15.77, and 208Pb/204Pb from ~38.95 to 39.38 (see 686 687 371 Peccerillo 2017, Chapter 7 and references therein). The isotopic data document a general 688 372 increase in 87Sr/86Sr and decrease in 143Nd/144Nd and 206Pb/204Pb from older to younger 689 87 86 143 144 206 204 690 373 rocks. However, the same Sr/ Sr increase and Nd/ Nd and Pb/ Pb decrease is 691 374 noted with increasing MgO content (Suppl. Fig. 1b). The isotopic variability indicates 692 693 375 that simple fractional crystallization cannot be the only process responsible for the 694 376 geochemical trends defined by rocks of the CF volcanic field. Pappalardo et al. (2002) 695 696 377 interpret the correlation between 87Sr/86Sr and age to reflect crustal contamination. Thus, 697 698 378 the younger erupted magmas would have resided in the crust for a longer time, underwent 699 379 more extensive crustal contamination and, therefore, show higher 87Sr/86Sr. However, it is 700 701 380 important to note that the age - isotopic signature correlation is not observed if only the 702 381 NYT and post-NYT bulk rock data are examined (Di Renzo et al., 2011). It is also worth 703 704 382 noting that the NYT shows homogeneous Sr-isotope ratios, and based on this Gebauer et 705 706 383 al. (2014) suggest minimal wall rock contamination for NYT magmas. In contrast with 707 384 the interpretation of Pappalardo et al. (2002), Peccerillo (2017) argues that the correlation 708 709 385 between isotopic composition and MgO may suggest that hotter and less viscous mafic 710 386 magmas experience more contamination relative to more evolved, cooler and more 711 712 387 viscous magmas, as has been invoked for Alicudi volcano in the Aeolian arc (Peccerillo 713 714 388 et al. 2017 and references therein). According to this latter interpretation, the isotopic 715 389 variability is not necessarily controlled by residence time of magmas in the crust. 716 717 390 Although most of the major element trends at IP suggest simple fractional 718 391 crystallization, variability of incompatible elements and variability of isotopic 719 720 392 composition of bulk rocks suggest that open-system magma processes must have also 721 722 393 been involved. The most mafic rocks in the Campanian Province are found in the IP, with 723 394 few of the lithic lava clasts considered to be primitive (MgO up to ~11.5 wt% and Ni up 724 725 726 13 727 728 729 730 731 395 to ~230 ppm; D'Antonio et al., 1999a). In addition, a juvenile sample from the Solchiaro I 732 733 396 deposits at IP indicates that lower Sr-isotopic ratios in the CF and IP (~0.7052) are not 734 87 86 735 397 restricted to lithic lava clasts (De Astis et al., 2004). The Sr/ Sr (0.7051 to 0.7065) of 736 398 IP bulk rocks is lower than that of CF rocks, and vice versa for 143Nd/144Nd. D'Antonio et 737 11 206 204 87 86 738 399 al. (2007) recognize two trends based on B and Pb/ Pb versus Sr/ Sr systematics 739 400 of bulk rocks from CF, IP, and Ischia Island. They interpret one trend to represent 740 741 401 contamination of the mantle source by slab-derived fluids or sediment-derived melts. In 742 743 402 particular, the CF rocks are interpreted to have been generated by a transitional MORB- 744 403 type asthenospheric mantle contaminated by sediment-derived melts, while the IP rocks 745 746 404 were generated by the same mantle source contaminated by slab-derived fluid. The other 747 405 trend records crustal contamination during crystallization of volcanic rocks younger than 748 749 406 39 ka. In addition, D'Antonio et al. (2007) suggest that some rocks younger than 39 ka 750 407 were affected by mixing-mingling processes between chemically and isotopically distinct 751 752 408 magmas, based on disequilibria among mineral and groundmass. Mazzeo et al. (2014) 753 754 409 also studied lithic lava clasts from the Solchiaro deposits at IP and hypothesized that the 755 410 mantle wedge from which primary magmas at IP were sourced was contaminated by 2- 756 757 411 4% of slab-derived components represented by partial melts from shale and carbonates, 758 412 and aqueous fluids from the slab. It is important to note that the mafic rocks from CF are 759 760 413 not primitive. Consequently, Pappalardo et al. (2002) and Peccerillo (2017) emphasize 761 762 414 that an interpretation whereby the difference in isotopic composition between CF and IP 763 415 is related to different types of contamination of the source is speculative. 764 765 766 416 Sample description 767 768 769 417 Six new samples were collected to supplement existing data from the two 770 418 volcanic areas. Five are from CF and one is from the IP. The CF samples are 771 772 419 representative of five volcanic units erupted at different times in this volcanic field. The 773 420 six eruptions sampled are, from oldest to youngest: Fossa Lupara (also named Senga in 774 775 421 the literature), Solfatara, Agnano-Monte Spina, Bacoli, and Neapolitan Yellow Tuff for 776 777 422 CF, and Solchiaro from the IP (Table 1 and Appendix Table A). 778 423 The Fossa Lupara sample (RESE17: Fig. 1) consists of black, sharp-edged scoriae 779 780 424 < 20 mm. The Fossa Lupara eruption has been dated at 3.82 ka by Di Vito et al. (1999), 781 782 14 783 784 785 786 787 425 and Smith et al. (2011) recalculated the age to be 3.98-4.20 ka using a more recently 788 789 426 developed and more precise method to interpret the isotopic data. The scoriae were 790 791 427 selected from Fossa Lupara deposits ejected during the intermediate stages of the 792 428 eruption (Lirer et al., 2011). The scoriae are porphyritic with abundant phenocrysts of 793 794 429 clinopyroxene and subordinate sanidine and were sampled at the northeast sector of the 795 430 crater associated with this eruption. 796 797 431 One sample representative of the upper deposits of the Solfatara eruption was 798 799 432 collected close to the northwest crater rim of Solfatara (RESO8: Fig. 1), and consists of 800 433 centimeter-scale fragments of rounded pumice contained in pyroclastic brownish ash 801 802 434 layers. The pumices contain abundant sanidine and clinopyroxene. The age of the 803 435 Solfatara eruption was recalculated from earlier data to be 4.15-4.35 ka by Smith et al. 804 805 436 (2011) based on chronological data reported by Di Vito et al. (1999) and by Isaia et al. 806 437 (2009). 807 808 438 The Agnano-Monte Spina sample (REMS12: Fig. 1) was collected ~ 200 m 809 810 439 southwest from the top of Monte Spina. The sample consists of sharp-edged pinkish-grey 811 440 pumices representing a 30-cm thick layer. The pumices are well sorted, characteristic of a 812 813 441 fallout deposit. This layer of pumice also shows reverse grading suggesting that some 814 442 changes occurred during the deposition such us (1) progressive increase in initial gas 815 816 443 velocity, (2) change in vent morphology or, (3) increase in eruption column density (Self, 817 818 444 1976 and references therein). Abundant sanidine, clinopyroxene, rare biotite and olivine 819 445 were observed. 820 821 446 The Bacoli sample (REBA1: Fig. 1) was collected close to the northern rim of the 822 447 crater of the Bacoli volcanic formation in the western rim of CF caldera (12.8 ka, Di 823 824 448 Renzo et al., 2011; 8.58 ka, Fedele et al., 2011; 11.51-14.15 ka, Smith et al., 2011). The 825 826 449 sample consists of a yellow tuff containing abundant isolated euhedral sanidine and 827 450 clinopyroxene phenocrysts. 828 829 451 The Neapolitan Yellow Tuff sample (14.9 ka; Deino et al., 2004) was collected 830 452 from the eastern caldera rim (northwestern part of Naples urban area, Ponti Rossi; 831 832 453 REPR11: Fig. 1). The sample consists of ash fall deposits with abundant sanidine and 833 834 454 clinopyroxene phenocrysts representing the unlithified facies of the NYT (this deposit is 835 455 referred to as Pozzolana, the material used for concrete production in Roman times). We 836 837 838 15 839 840 841 842 843 456 selected only single phenocrysts showing euhedral habit and rim glass to avoid including 844 845 457 xenocrysts in our sample suite. 846 847 458 The Solchiaro sample (RESC2: Fig. 1) was collected on the northeastern coast of 848 459 IP and represents the distal unlithified facies of the Solchiaro deposits. The sample 849 850 460 consists of grey ash, brown lapilli, and isolated olivine, clinopyroxene, and sanidine 851 461 crystals [sample RESC2 in Esposito et al. (2011)]. 852 853 854 462 Analytical methods 855 856 857 463 The samples selected for this study were gently crushed and individual phenocrysts were 858 464 hand-picked from the crushed material under a binocular microscope. Each phenocryst 859 860 465 was mounted on a 2.5 mm-diameter glass rod (Esposito et al., 2014; Thomas and Bodnar, 861 466 2002) using Crystalbond™ cement. Phenocrysts were polished on the top and bottom 862 863 467 surfaces and those crystals containing rounded to regular-shaped MI were selected for 864 865 468 study. After we selected 20-30 crystals (depending on size), the phenocrysts were pressed 866 469 into a one-inch diameter indium mount. Indium was used as the mounting medium to 867 868 470 prevent H-C contamination during secondary ion mass spectrometric (SIMS) analysis, as 869 471 occurs when epoxy mounts are used. 870 871 472 MI in the Fossa Lupara sample and one MI hosted in sanidine from the Agnano 872 873 473 Monte Spina sample were partially to completely crystallized as found. We performed 874 474 heating experiments on the crystallized MI using a Vernadsky heating stage (Sobolev et 875 876 475 al., 1980). The heating stage was calibrated according to the 1-atm melting temperatures 877 476 of silver (962 °C) and gold (1064 °C). During the heating experiment, a continuous flow 878 879 477 of He into the furnace of the heating stage prevented (or minimized) Fe-oxide formation 880 881 478 on the surface of Fe-rich phenocrysts. MI were heated until the T of homogenization (MI 882 479 containing only silicate melt) was reached. Following homogenization, the sample was 883 884 480 quenched rapidly to room temperature. The T uncertainty is ~ ±5ºC at 1064 °C. Heating 885 481 rates were varied during the run, depending on phase behavior of the MI (Esposito et al., 886 887 482 2012). 888 889 483 The major element compositions of glasses and minerals were determined first 890 484 using a JEOL JXA 8900 electron microprobe (EMP) at the USGS (Reston, VA, USA). 891 892 485 The analyses were conducted using an accelerating voltage of 15 kV. The beam current 893 894 16 895 896 897 898 899 486 was 10 nA and a defocused beam of 10 µm diameter was used. For oxide concentrations 900 901 487 > 1 wt%, 1-sigma errors are < 2% relative, while for oxide concentrations < 1 wt%, 1 902 903 488 sigma errors are generally < 10% relative. Additional information relative to EMP 904 489 standards used for the analyses are included in Supplementary Table B. 905 906 490 Following EMP analyses, volatile elements (O, H, C, F, S, Cl) were measured by 907 491 SIMS (Cameca IMS 1280) at the Woods Hole Oceanographic Institution (Woods Hole, 908 909 492 MA, USA). We performed the analyses using 133Cs+ as the ion source and the current 910 911 493 varied between 1.0 and 1.6 nA. For each spot analyzed, the surface of the sample was 912 494 pre-sputtered for 240 seconds to remove surface contamination. The size of the rastered 913 914 495 area was 30 µm x 30 µm, and a 15 µm x 15 µm spot in the center of the rastered area was 915 496 analyzed from 10 to 15 times in depth profile mode. We tuned the SIMS to count ions 916 917 497 related to masses 16O, 1H, 12C, 19F, 32S, 35Cl, and 30Si. Ratios between ion counts per unit 918 498 time of each mass measured and ion counts per unit time of 30Si were used to quantify the 919 920 499 volatile contents. Six basaltic to basaltic-andesitic glass standards were used for 921 922 500 calibration of the SIMS, and additional information on the glass standards is reported by 923 501 Helo et al. (2011). For the SIMS measurements, using glass standards with silicate 924 925 502 compositions that differ from MI compositions may introduce a matrix effect. However, 926 503 Hauri et al. (2002) showed that SIMS calibration curves for H O, CO , F, S, and Cl 927 2 2 928 504 obtained using rhyolitic, andesitic, and basaltic glass standards do not show significant 929 930 505 differences below certain absolute concentrations – specifically, for H2O contents <2 931 506 wt% and CO2 contents < 1000 ppm. Most of the volatile concentrations of the MI 932 933 507 reported in this study are below these upper limits. We used the calibration curves of 934 508 Esposito et al. (2011) to calculate volatile concentrations in MI. To avoid contamination 935 936 509 from the C-coating used for earlier EMP analysis, we adopted a specific protocol (see 937 938 510 also discussion by Esposito et al. 2014) to remove the C coating. In particular, ~5 µm of 939 511 sample was removed by polishing following EMP and before SIMS analysis, and the 940 941 512 SIMS analytical area was pre-sputtered for 290 s before data collection started. We 942 513 selected only flat depth profiles whereby the C counts did not vary significantly with 943 944 514 depth, and we continuously monitored the ion-count response image to confirm a 945 946 515 homogeneous carbon signal across the rastered area during analysis. Relative precision 947 948 949 950 17 951 952 953 954 955 516 for CO , H O, F, Cl, and S by SIMS is <10% relative, based on repeated analysis of glass 956 2 2 957 517 standards (see Esposito et al, 2014; their Figure 14). 958 959 960 518 Results 961 962 519 Petrography of melt inclusions. Many (but not all) of the phenocrysts examined 963 964 520 contain MI and other mineral inclusions. Fluid inclusions are rare in the phenocrysts 965 521 studied. Various types of MI, based on phases present at room temperature, were 966 967 522 observed in all five samples as described below. 968 969 523 Sanidine, clinopyroxene, and plagioclase from Bacoli, Solfatara, Solchiaro, and 970 524 NYT samples contain MI filled with mostly glass. Most of these naturally glassy MI 971 972 525 contain one bubble in addition to the glass phase, and some contain glass plus bubble(s) 973 526 plus an associated mineral (most commonly apatite). In a few cases, naturally glassy MI 974 975 527 contain only glass (bubble-free MI). MI containing more than one bubble are found 976 977 528 mostly in sanidine phenocrysts. This is thought to represent bubble formation in the MI 978 529 after the melt has become sufficiently viscous to prevent the multiple bubbles from 979 980 530 coalescing into a single bubble. Bubble-bearing MI (glass plus one bubble) show similar 981 531 volume fractions of vapor (i.e., volume /volume ) in single phenocrysts selected for 982 bubble MI 983 532 our study, and the proportion of vapor never exceeds 5 volume %, suggesting that the 984 985 533 vapor bubble nucleated after trapping of a homogeneous silicate liquid (Moore et al., 986 534 2015), i.e., vapor bubbles were not trapped along with the melt. It is important to note 987 988 535 that NYT MI show a larger proportion of bubble-free MI relative to MI from the other CF 989 536 and IP eruptions studied. 990 991 537 The phenocrysts selected from Agnano-Monte Spina and Fossa Lupara samples 992 993 538 contain partially to totally crystallized MI, in addition to naturally glassy MI as described 994 539 above. In particular, MI in sanidine from Agnano-Monte Spina are sometimes found 995 996 540 partially or totally crystallized, while partially crystallized MI are rarely found in 997 541 clinopyroxene. Most of the Fossa Lupara phenocrysts contain partially to totally 998 999 542 crystallized MI, but sometimes sanidine phenocrysts contain naturally glassy MI. Some 1000 1001 543 sanidine phenocrysts in the Fossa Lupara sample show decrepitation halos, suggesting 1002 544 that the MI developed high fluid pressures after trapping and have lost some of their 1003 1004 545 volatiles via fracturing of the surrounding host (Cannatelli et al., 2016). 1005 1006 18 1007 1008 1009 1010 1011 546 In general, MI in all samples are ovoid or negative-crystal shape, but sometimes 1012 1013 547 slightly to highly irregular shapes were observed (Table 1). For this study, we selected 1014 1015 548 regularly shaped MI because irregularly-shaped MI are more likely to have leaked or 1016 549 were in communication with the melt adjacent to the growing crystal for a prolonged 1017 1018 550 period of time, (Humphreys et al., 2008a; Humphreys et al., 2008b; Welsch et al., 2013) 1019 551 or represent MI that have decrepitated (Maclennan, 2017). In addition, irregularly shaped 1020 1021 552 MI hosted in plagioclase may be the result of dissolution reactions and record a chemical 1022 1023 553 composition that is not representative of the melt in equilibrium with the host (Nakamura 1024 554 and Shimakita, 1998). Size ranges from a few microns to ~300 µm considering the 1025 1026 555 longest dimension, and we selected only MI > 20 µm to assure that analytical spots for 1027 556 EMP and SIMS would be contained entirely within the MI, thus avoiding contributions 1028 1029 557 from the surrounding host. Most of the MI are isolated in the cores of phenocrysts 1030 558 without any geometrical correlation with their host (i.e., the MI are not obviously aligned 1031 1032 559 along crystal growth surfaces). Some phenocrysts contain more than one MI. Consistent 1033 1034 560 with the results of Esposito et al. (2014), MIA are rarely observed; where they are, 1035 561 individual MI are < 20 µm, a size that is smaller than the optimal instrumental conditions 1036 1037 562 for analyzing major elements and volatiles. 1038 563 Microthermometry. As mentioned above, most of the Fossa Lupara MI were 1039 1040 564 crystallized as found. For this reason, we performed heating experiments using a 1041 1042 565 Vernadsky heating stage to produce a homogenous glass before analysis. All MI-bearing 1043 566 clinopyroxenes from Fossa Lupara show a homogeneous silicate melt phase upon heating 1044 1045 567 to ~ 1200ºC and all were quenched at this temperature. Most MI contain only glass after 1046 568 quenching from ~1200ºC, but some MI developed a vapor bubble during quenching. MI 1047 1048 569 in sanidine phenocrysts from Fossa Lupara were quenched from ~1070ºC and most MI 1049 1050 570 show only glass after quenching. 1051 571 Major element compositions. Major element concentrations of MI are plotted on 1052 1053 572 Harker diagrams versus SiO2 (Fig. 2). The major element compositions of MI from this 1054 573 study (Table 2) overlap with published MI compositions (Arienzo et al., 2016; Arienzo et 1055 1056 574 al., 2010; Cannatelli et al., 2007; Cecchetti et al., 2002-2003; Cipriani et al., 2008; 1057 1058 575 Esposito et al., 2011; Fourmentraux et al., 2012; Mangiacapra et al., 2008; Mormone et 1059 576 al., 2011; Roach, 2005; Stock et al., 2016) and with bulk rock compositions from the 1060 1061 1062 19 1063 1064 1065 1066 1067 577 literature (Albini et al., 1977; Beccaluva et al., 1991; Brocchini et al., 2001; Cannatelli et 1068 1069 578 al., 2007; Capaldi et al., 1972; D'Antonio et al., 1999a; D'Antonio et al., 1999b; De Astis 1070 1071 579 et al., 2004; de Vita et al., 1999; Di Girolamo et al., 1984; Di Vito et al., 2011; Ghiara et 1072 580 al., 1979; Mastrolorenzo and Pappalardo, 2006; Orsi et al., 1995; Orsi et al., 1992; 1073 1074 581 Pappalardo et al., 2002; Piochi et al., 2005; Piochi et al., 2008; Scarpati et al., 1993; 1075 582 Tonarini et al., 2009; Tonarini et al., 2004; Turi et al., 1991; Villemant, 1988; Wohletz et 1076 1077 583 al., 1995) (Fig. 2). 1078 1079 584 Compositions of MI studied here define differentiation trends with higher degrees 1080 585 of scatter relative to trends defined by bulk rock compositions – this is similar to trends 1081 1082 586 defined by data for MI from the literature. In particular, at the same SiO2 concentration, 1083 587 some major element concentrations (e.g., K2O and Na2O) span a wider range compared to 1084 1085 588 that for the bulk rocks (Fig. 2). This behavior is commonly observed in MI studies (Kent, 1086 589 2008, and references therein). In general, the major element concentrations of MI plotted 1087

1088 590 as a function of SiO2 show continuous differentiation trends (Fig. 2 a, b, d, and e). MI 1089 1090 591 compositions from this study mostly agree with differentiation trends for bulk rocks, 1091 592 representing mainly fractional crystallization dominated by olivine plus clinopyroxene 1092 1093 593 crystallization for less evolved magmas, clinopyroxene for intermediate magmas, and 1094 594 sanidine plus clinopyroxene for more evolved magmas. 1095

1096 595 Differentiation trends for TiO2, FeOtot, and P2O5 (Fig. 2), as well as for Al2O3, 1097 1098 596 show poorer agreement compared to trends defined by bulk rock compositions from the 1099 597 literature. This behavior is consistent with MI data reported in the literature for other 1100 1101 598 magmatic systems. For instance, TiO2 contents of MI from this study show weak 1102 599 correlations with crystallization indicators relative to bulk rock data from the literature 1103 1104 600 (Fig. 2d). For the least evolved MI compositions, TiO2 concentrations span a wide range, 1105 1106 601 while for more evolved compositions TiO2 concentrations are more restricted. 1107 602 For the MI studied here, FeOtot behaves in a manner similar to TiO2 (Fig. 2e). In 1108 1109 603 particular, less evolved MI show more scatter relative to the more evolved MI. Some MI 1110 604 from the less evolved group (SiO concentration ranging from ~47 to ~54 wt%) are FeO- 1111 2 1112 605 depleted relative to the bulk rocks. The Fe-poor MI are all hosted in clinopyroxene from 1113 1114 606 the Fossa Lupara sample. Some MI data from the literature for the Minopoli1 and Fondo 1115 607 Riccio eruptions (Cannatelli et al., 2007) and for the Solchiaro eruption (Mormone et al., 1116 1117 1118 20 1119 1120 1121 1122 1123 608 2011) show the same behavior, and most of the MI are hosted in clinopyroxene (three MI 1124 1125 609 are hosted in olivine). In addition, a group of MI hosted in olivine and clinopyroxene 1126 1127 610 from the Minopoli2 and Fondo Riccio eruptions (Mangiacapra et al., 2008) show 1128 611 extremely low FeOtot contents at intermediate SiO2 contents. Also, some MI hosted in 1129 1130 612 clinopyroxene from the Astroni1 eruption (Stock et al., 2016) show a depletion in FeOtot 1131 613 relative to the bulk rock trend. The depletion of FeO may reflect Fe loss from the MI to 1132 1133 614 the Fe-Mg host via diffusion, as has been proposed and documented for other volcanic 1134 1135 615 systems (Danyushevsky et al., 2000). It is important to note that most of the Fe-poor and 1136 616 less evolved MI from CF were heated experimentally. Some of the Fe-poor MI also show 1137 1138 617 slight depletion in Al2O3, slightly higher CaO contents, lower Na2O contents, and slightly 1139 618 higher K2O (Fig. 2). 1140 1141 619 Several evolved MI from this study show a slight depletion of Na2O relative to the 1142 620 bulk rock geochemical trends, with the exception of MI from the Solfatara and Astroni 1143 1144 621 samples. Some of the bulk rock compositions show the same Na-depletion, and several 1145 1146 622 MI compositions reported in the literature show the same behavior. Na2O-poor MI are not 1147 623 preferentially hosted in sanidine or clinopyroxene phenocrysts, and they include both 1148 1149 624 heated and naturally quenched MI. Na2O-poor MI can indicate that Na has been lost from 1150 625 the MI by diffusion, and to test this hypothesis we calculated totals considering both 1151 1152 626 EMP and SIMS concentrations. Some of the Na-poor MI show ~ 100 wt% totals (Table 1153 1154 627 2), suggesting that Na loss is minimal. Also, there is no correlation between Na2O and 1155 628 H2O concentrations. In addition, the group of MI showing depletion of Na2O shows a 1156 1157 629 slight enrichment in K2O. Potassium should not increase as a result of Na loss by 1158 630 diffusion (Morgan and London, 2005). It is also important to note that Na-loss should not 1159 1160 631 be significant at the conditions used to analyze major element concentrations (Hayward, 1161 1162 632 2012). The least evolved MI from this study, as well as those from the literature, show a 1163 633 wide range of K2O concentrations. For instance, for MI with SiO2 of ~50 wt%, K2O 1164 1165 634 varies by a factor of 6 (from ~1 to ~6 wt%) (Fig. 2b). 1166 635 Among major and minor elements, P O shows the poorest correlation with SiO 1167 2 5 2 1168 636 content. P2O5 concentrations of mafic and intermediate composition MI both from this 1169 1170 637 study and from the literature show considerable scatter (from ~45 to ~57 wt% SiO2, see 1171 1172 1173 1174 21 1175 1176 1177 1178 1179 638 Fig. 2 f), while evolved MI (from ~57 to ~66 wt% SiO ) show good correlations between 1180 2 1181 639 P2O5 contents and crystallization indicators. 1182 1183 640 A group of MI hosted in plagioclase from the Bacoli eruption and one bulk rock 1184 641 composition from the literature do not overlap with the general trends defined by MI 1185 1186 642 from the literature, or with bulk rock data for most of the major elements. The 1187 643 compositions of plagioclase-hosted MI are Qz normative, an observation that is rare for 1188 1189 644 bulk rocks of CF and IP. These plagioclase-hosted MI show Na2O and K2O 1190 1191 645 concentrations consistent with CF and IP bulk rock differentiation trends, while Al2O3 1192 646 and SiO2 concentrations are slightly higher than the bulk rock differentiation trends (Fig. 1193 1194 647 2). Concentrations of TiO2, CaO, FeO, and P2O5 are lower than those shown by the bulk 1195 648 rock differentiation trends. The compositions of the MI hosted in plagioclase are not 1196 1197 649 representative of CF and IP magmatism, suggesting that plagioclase selected for the study 1198 650 represent xenocrysts or that they formed under disequilibrium conditions. 1199 1200 651 Finally, we note that some of the individual bulk rock compositions project off the 1201 1202 652 main bulk-rock trend (Fig. 2), and these tend to lie along mixing lines that point toward 1203 653 clinopyroxene or feldspar compositions. This suggests that some of the bulk rock 1204 1205 654 compositions represent mixtures of melt plus entrained crystals. These “anomalous” bulk 1206 655 rock compositions represent a small fraction of the complete dataset. 1207

1208 656 Volatiles. Among all volatiles analyzed, Cl shows the best correlation with SiO2 1209 1210 657 concentration, especially for less-evolved melts, while S shows a weak correlation, and 1211 658 H2O, CO2, and F show no correlation with SiO2 (Fig. 3). Considering MI hosted in all 1212 1213 659 phases, we observed two important features: (1) volatile concentrations of the MI studied 1214 660 here overlap with volatile concentrations of MI from CF and IP reported in the literature 1215 1216 661 and (2) volatile concentrations of MI from this study are higher relative to CF and IP bulk 1217 1218 662 rocks reported in the literature. This latter observation highlights the limitations of using 1219 663 bulk rocks to determine volatile compositions of magmas, because these samples have 1220 1221 664 partially or totally degassed during ascent and eruption onto the surface. Conversely, MI 1222 665 are more likely to retain the original pre-eruptive volatile content during ascent and 1223 1224 666 eruption. 1225 1226 667 Chlorine concentrations range from 913 to 7931 ppm. Chlorine behaves 1227 668 incompatibly through most of the magmatic differentiation showing a good correlation 1228 1229 1230 22 1231 1232 1233 1234 1235 669 with SiO <60 wt% (0.85 r2), but for SiO >60 wt% Cl shows a sub-vertical trend from 1236 2 2 1237 670 7710 to 1188 ppm (Fig. 3a). This vertical trend is observed for all MI hosted in sanidine. 1238 1239 671 Sulfur concentrations range from below detection limit (bdl) to 2884 ppm. Sulfur 1240 672 generally decreases from less evolved MI to more evolved MI. The S contents of MI are 1241 1242 673 more scattered for less evolved MI, while S is more clustered for more evolved 1243 674 compositions (Fig. 3b). 1244 1245 675 Concentrations of H2O in the MI studied here range from bdl to 2.54 wt%. In 1246 1247 676 contrast to Cl and S, H2O shows poor correlations with SiO2, but for MI with >60 wt% 1248 677 SiO2 a sub-vertical distribution is observed (Fig. 3d). It is important to recognize that the 1249 1250 678 H2O-SiO2 trend for MI with <60 SiO2 is defined by only five MI. 1251 679 Fluorine concentrations range from 81 to 3860 ppm. Fluorine concentrations of 1252 1253 680 MI show a behavior similar to Cl, but the correlation is weaker. For MI >60 SiO2 the 1254 681 distribution is sub-vertical as observed for Cl and H O. Since there are only five MI 1255 2

1256 682 showing SiO2 <60 wt%, the apparent vertical trend may reflect the small sample size 1257 1258 683 (Fig. 3e). 1259 684 Carbon dioxide concentrations range from below detection limit to 258 ppm with 1260 1261 685 one MI from the NYT sample showing 1046 ppm (Fig. 3c). Carbon dioxide does not 1262 686 correlate with SiO even when a single eruption is considered (Fig. 3c), and H O-CO 1263 2 2 2 1264 687 systematics do not define a trend (Fig. 4a). In contrast, MI from individual eruptions are 1265 1266 688 considered, the NYT MI show a positive trend (Fig. 4b). The correlation shown by NYT 1267 689 MI should be viewed with caution because only three analyses are available. 1268 1269 690 In general, volatile abundances of MI do not correlate with each other, with the 1270 691 exception of Cl vs F (Fig. 4c) which shows a strong positive correlation. This suggests 1271 1272 692 that Cl and F are retained in the melt during the initial and intermediate stages of 1273 1274 693 differentiation (trachybasalt-shoshonite-latite). During the last stages of crystallization, Cl 1275 694 likely partitions strongly into the fluid phase while F is retained in the melt. 1276 1277 695 Host mineral compositions. MI from this study are hosted in clinopyroxene (cpx), 1278 696 sanidine, and plagioclase. The ferrosilite component of cpx hosting MI ranges from 6 to 1279 1280 697 18 mol%, the wollastonite component from 46 to 49 mol%, and the enstatite component 1281 1282 698 ranges from 34 to 48 mol% (Suppl. Fig. 2 c). The MI compositions and host phase 1283 699 compositions are both consistent with trapping of a melt that was in equilibrium with cpx 1284 1285 1286 23 1287 1288 1289 1290 1291 700 during crystallization. For example, the SiO concentration of MI correlates with the 1292 2 1293 701 ferrosilite component. Note that MI data show some significant scatter, and we will 1294 1295 702 discuss the reasons for this in the discussion section. MI and clinopyroxene host 1296 703 compositions are consistent with data reported in the literature. 1297 1298 704 The sanidines containing MI analyzed in this study possess 54 to 86 mol% 1299 705 orthoclase component, with most in the range 76-86 mol% (Suppl. Fig. 2 a and b). The 1300 1301 706 Or-poor sanidines are restricted to the Solchiaro sample and show up to 42 mol% albite 1302 1303 707 component. No correlation is observed between host composition and SiO2 concentration 1304 708 of MI (Suppl. Fig. 2 a). One possible explanation is that sanidine began to crystallize over 1305 1306 709 a restricted interval of SiO2 concentrations (see discussion section for more details). 1307 710 However, a good correlation is observed if the orthoclase component of the host is 1308 1309 711 plotted versus SiO2/CaO ratio of the MI (Suppl. Fig. 2 b). The SiO2/CaO ratio is likely a 1310 712 better crystallization indicator at this differentiation stage, and, also, this ratio is not 1311 1312 713 affected by PEC or host dissolution (see the discussion section for more details). The 1313 1314 714 orthoclase component decreases and the albite component increases as crystallization 1315 715 progresses. This behavior was reported and documented by Melluso et al. (2012) who 1316 1317 716 showed that the NaO content of the alkali feldspar increases in the more evolved bulk 1318 717 rocks. 1319 1320 718 Only six MI from this study and from the literature combined are hosted in 1321 1322 719 plagioclase. The anorthite component ranges from 75 to 81 mol%, and the anorthite 1323 720 component of the host correlates with the SiO2 concentration of the MI. This observation, 1324 1325 721 based on only six MI, should be taken with caution. 1326 722 Post entrapment crystallization or dissolution. In this study we did not attempt to 1327 1328 723 correct for changes in MI composition associated with post entrapment crystallization 1329 1330 724 (PEC) owing to the lack of a reliable model to implement the correction. We note that 1331 725 Kress and Ghiorso (2004) report that an algorithm incorporated into the MELTS platform 1332 1333 726 could be used to estimate PEC based on the composition of the MI and the host for 1334 727 various phenocrysts, including pyroxene and feldspar. We attempted to apply this method 1335 1336 728 to correct the MI of this study but the results indicated that ~ 90% PEC had to occur for 1337 1338 729 many MI. This value is not realistic and is not supported by petrographic examination of 1339 730 the MI. As such, we consider bubble-free MI having compositions that overlap with bulk 1340 1341 1342 24 1343 1344 1345 1346 1347 731 rock compositions to have experienced only minimal PEC, and have assumed that the 1348 1349 732 compositions of the MI reasonably approximate the original melt composition. 1350 1351 1352 733 Discussion 1353 1354 734 Our interpretation of geochemical trends in MI from CF and IP is based on new data from 1355 1356 735 this study as well as published data for CF and IP. First, we focus on major element 1357 736 concentrations of MI that do not follow the general trends defined by bulk rock 1358 1359 737 compositions. We test if the MI compositions are representative of the silicate melt in 1360 1361 738 equilibrium with the host mineral, based on major element compositions. We discuss the 1362 739 significance of anomalous MI, and we exclude the non-representative anomalous MI 1363 1364 740 from the interpretation of volatile evolution at CF and IP. Second, we focus on volatile 1365 741 and major element concentrations of normal MI that have trapped and preserved a 1366 1367 742 representative sample of the melt at the time of trapping. We discuss the significance of 1368 1369 743 volatile concentrations of all MI, and we select only the bubble-free MI to investigate 1370 744 crystallization trends using the rhyolite-MELTS code (Ghiorso and Gualda, 2015). 1371 1372 1373 745 Significance of Anomalous MI 1374 746 Anomalous melt inclusions are those whose compositions cannot be reasonably inferred 1375 1376 747 to represent compositions of melts along a fractionation trend. Below we discuss the 1377 1378 748 anomalous MI that were identified and possible processes that may have produced the 1379 749 anomalous compositions. 1380 1381 750 1382 751 Al2O3 and P2O5 anomalies of MI 1383 1384 752 Several processes can lead to anomalous Al2O3 and P2O5 concentrations in MI. These 1385 753 anomalous compositions can represent the trapping of a boundary layer associated with 1386 1387 754 rapid crystal growth rates (e.g., Baker, 2008), or could represent trapping of melt 1388 1389 755 produced as a result of dissolution-reaction-mixing (DRM) processes in the mush zone 1390 756 (Danyushevsky et al., 2004). Alternatively, these anomalies can be artifacts generated as 1391 1392 757 a result of overheating during laboratory heating experiments. In this section, we discuss 1393 758 which processes are likely responsible for the anomalous MI observed at CF and IP. 1394 1395 1396 1397 1398 25 1399 1400 1401 1402 1403 759 As reported above, several MI are enriched in Al O relative to bulk rocks in 1404 2 3 1405 760 differentiation plots. Including MI from the literature, we observed that Al2O3 is one of 1406 1407 761 the major components that shows large deviations from bulk rock geochemical trends. MI 1408 762 hosted in clinopyroxene show the largest variations for Al2O3 (Fig. 5a). At intermediate 1409 1410 763 CaO concentrations (between 5 and 8 wt%), some MI hosted in clinopyroxene show 1411 764 Al2O3 contents that are elevated relative to bulk rocks. At high CaO concentrations, other 1412 1413 765 MI hosted in clinopyroxene shows anomalously low Al2O3 concentrations relative to the 1414 1415 766 bulk rocks. It is worth noting that a few MI hosted in olivine show a slight enrichment in 1416 767 Al2O3 content. Experimental studies suggest that host-incompatible element 1417 1418 768 concentrations of MI could be affected by boundary layer process. In particular, Faure 1419 769 and Schiano (2005) reported that MI trapped in fast growing olivine show a significant 1420 1421 770 enrichment in Al concentration. However, boundary layer effects alone cannot explain 1422 771 the CaO vs. Al O deviations of MI from the bulk rock trends. Interestingly, the Al O - 1423 2 3 2 3

1424 772 poor MI were all reheated in the laboratory, while the most of the Al2O3-rich MI are 1425 1426 773 naturally quenched (Fig. 5b). These Al2O3-rich MI show low TiO2, FeOtot, and MgO, and 1427 774 this observation is consistent with high degrees of PEC or with a plagioclase component 1428 1429 775 added to the “normal” CF melt composition. However, it is important to note that most of 1430 776 the Al O -rich MI also show elevated P O and some show high Cl concentrations, 1431 2 3 2 5 1432 777 consistent either with trapping of apatite along with the melt or that some MI hosted in 1433 1434 778 clinopyroxene trapped a boundary layer melt that was enriched in P (Baker, 2008). 1435 779 For most of the Al-rich MI, the P2O5 enrichment cannot be attributed to PEC. If 1436 1437 780 we consider the Al-rich MI showing the highest P2O5 concentration [1.38 wt%, “FRC2- 1438 781 P7-M1” from Cannatelli et al. (2007), see database Appendix Table D] and the BR 1439 1440 782 showing the highest P2O5 concentration (0.6 wt%, “s 97110PNA-S”), ~ 57% PEC would 1441 1442 783 be required to bring the MI composition into coincidence with the BR trend. With this 1443 784 amount of PEC other phases in addition to the host would also be crystallizing in the MI. 1444 1445 785 Also, the Al-rich MI under consideration was reheated in the lab, and was quenched 1446 786 when the MI homogenized, indicating only minor PEC if the quenching occurred from a 1447 1448 787 temperature that was lower than the trapping temperature. Also, at 57% PEC the Al2O3 1449 1450 788 concentration should have been ~11 wt%, a value that is much lower than the BR with 1451 789 similar extent of differentiation. To explain the P2O5 enrichment of FRC2-P7-M1 only 1452 1453 1454 26 1455 1456 1457 1458 1459 790 requires ~2% apatite component, while the enrichment of Al O would require a ~23% 1460 2 3 1461 791 plagioclase component. 1462 1463 792 It is generally assumed that boundary layer processes would affect compositions 1464 793 of smaller inclusions to a greater extent compared to larger inclusions being trapped at 1465 1466 794 the same crystal-melt interface. Esposito et al. (2011) found no correlation between the 1467 795 MI size and the P2O5 and K2O concentration, suggesting that boundary layer processes 1468 1469 796 alone cannot explain the Al2O3-P2O5 MI trends. Furthermore, an inverse correlation 1470 1471 797 between the size of MI and the concentration of slow diffusing elements should be 1472 798 expected only if there is evidence that all the MI being considered trapped the same melt. 1473 1474 799 In other words, we cannot compare data for a group of MI if there is no evidence that 1475 800 each MI was originally trapped from the same bulk melt. This fact underscores the 1476 1477 801 importance of studying MIA (Bodnar and Student, 2006) to test for correlations (if any) 1478 802 between MI composition and size. With the assumption that all MI comprising the MIA 1479 1480 803 trapped the same melt, the correlation between size and concentration of slow diffusing 1481 1482 804 elements can be used to test whether boundary layer processes significantly affected the 1483 805 composition of those MI, and corrections to determine the original trapped melt 1484 1485 806 composition can be proposed. 1486 807 In this study, all MI hosted in sanidine show similar compositions and show no 1487

1488 808 correlation between size of the MI and the K2O concentration (Suppl. Fig. 4), suggesting 1489 1490 809 that boundary layer processes did not affect the composition of sanidine-hosted MI. The 1491 810 selective enrichment of major and trace elements can be interpreted as the result of 1492 1493 811 dissolution-reaction-mixing (DRM) in a mush zone at the interface between a magma 1494 812 body and wall rock (Danyushevsky et al., 2004; Esposito et al., 2011). Thus, the 1495 1496 813 enrichment of Al2O3 can be explained by DRM between plagioclase and melts at various 1497 1498 814 stages of evolution. Combined enrichment of Al2O3 and P2O5 can represent DRM 1499 815 between less evolved melt and mush zone material of more evolved magmas rich in 1500 1501 816 plagioclase and apatite beneath CF and IP. The DRM interpretation implies micro-scale 1502 817 heterogeneity that results when a less evolved melt ascends and interacts with a 1503 1504 818 preexisting mush zone. In particular, the DRM interpretation corroborates the hypothesis 1505 1506 819 that basaltic and trachybasaltic magmas interacted with a mush zone consisting of 1507 1508 1509 1510 27 1511 1512 1513 1514 1515 820 cogenetic evolved latitic and trachytic-phonolitic magmas at CF and IP, as reported in 1516 1517 821 other studies (D'Antonio, 2011 and references therein). 1518 1519 822 As reported above, some of the heated and less evolved MI hosted in 1520 823 clinopyroxene show enrichment in CaO and MgO, and depletion in Al2O3, TiO2, and 1521 1522 824 Na2O. These anomalies indicate that MI have likely been over-heated in the laboratory 1523 825 and the resulting composition of MI shows dilution of clinopyroxene-incompatible 1524 1525 826 elements and enrichment in clinopyroxene-compatible elements (Fig 5b). We attempted 1526 1527 827 to correct compositions of reheated MI hosted in clinopyroxene by adjusting the CaO and 1528 828 Al2O3 to the regression line of the bulk rock data or to the bulk rock data “cloud” in the 1529 1530 829 CaO vs. Al2O3 plot (Suppl. Fig. 3a). This correction suggests that 30 to 33% of host 1531 830 dissolution would have occurred during heating. However, after this correction, FeO, 1532 1533 831 K2O and P2O5 still show anomalous concentrations (Suppl. Fig. 3 b, c, and d). In fact, a 1534 832 “correction” for 30-33% host dissolution tends to drive both K O and P O towards more 1535 2 2 5 1536 833 anomalous compositions. This suggests that a simple host dissolution correction cannot 1537 1538 834 be used to restore the original MI composition. 1539 835 Heated and more evolved MI hosted in sanidine show higher K2O concentrations 1540 1541 836 relative to unheated and more evolved MI (Suppl. Fig. 4), similar to MI hosted in cpx. MI 1542 837 showing anomalous compositions for some major elements may also show volatile 1543 1544 838 concentrations that are not representative of the silicate melt that was in equilibrium with 1545 1546 839 the host at the time of trapping. For instance, if the anomalous composition of a MI is the 1547 840 result of DRM, the MI volatile budget could be modified, and the concentrations of 1548 1549 841 individual volatile components could be differentially affected. If the wallrocks involved 1550 842 are dominantly composed of anhydrous phases, concentrations of all volatiles would be 1551 1552 843 lowered by simple dilution (Danyushevsky et al., 2004; Esposito et al., 2011). 1553 1554 844 Alternatively, if the wallrocks contain carbonates, as has been inferred for Vesuvius (Gilg 1555 845 et al., 2001) and Merapi (Deegan et al., 2010), the concentration of CO2 in the magma 1556 1557 846 could be increased by DRM processes. The eruptive products at CF lack carbonate- 1558 847 bearing xenoliths or skarn materials (D'Antonio, 2011). This suggests that DRM at CF 1559 1560 848 should lower the volatile contents of the melts, and this is consistent with observations by 1561 1562 849 Esposito et al. (2011) who reported very low S, CO2 and Cl concentrations of some 1563 850 anomalous MI and compositions plot towards the origin on S-CO2-Cl vs Zr binary 1564 1565 1566 28 1567 1568 1569 1570 1571 851 diagrams. The low concentration of volatiles is expected for overheated MI. Even though 1572 1573 852 the lowering of the volatile concentration caused by dissolving excess host during the 1574 1575 853 heating experiment is likely insignificant, two additional issues are associated with 1576 854 overheated MI. If MI are heated to a temperature above the “true” trapping temperature, 1577 1578 855 the total amount of time that the MI would have been exposed to “high” temperatures 1579 856 will be longer, compared to heating to a lower temperature at the same rate. This, in turn, 1580 1581 857 translates into a greater likelihood to lose H+ by diffusion (Bucholz et al., 2013; Mironov 1582 1583 858 and Portnyagin, 2011). Hydrogen can also be lost from the MI in nature (before 1584 859 eruption), leading to an increase in the homogenization temperature of the MI and 1585 1586 860 requiring heating to higher temperatures to achieve homogenization during re-heating 1587 861 experiments in the lab. 1588 1589 862 1590 863 Fe exchange 1591 1592 864 Numerous studies have documented Fe loss from MI based on the negative 1593 1594 865 correlation between FeOtot of the MI and the composition (Fo mol%) of the host, as well 1595 866 as the lower FeOtot in the MI relative to bulk rocks and matrix glasses of the same 1596 1597 867 magmatic system at comparable degrees of differentiation (Danyushevsky et al., 2000; 1598 868 Danyushevsky et al., 2002b; Esposito et al., 2011; Norman et al., 2002; Sobolev and 1599 1600 869 Danyushevsky, 1994; Yaxley et al., 2004). Six MI from this study show anomalously low 1601 1602 870 Fe concentrations relative to the main differentiation trend defined by bulk rocks (Fig. 2 1603 871 e). Three MI are hosted in plagioclase from the Bacoli eruption, and the other three MI 1604 1605 872 are hosted in clinopyroxene from the Fossa Lupara eruption, and these were heated in the 1606 873 laboratory to obtain a homogeneous glass. The Fe-poor MI hosted in plagioclase from 1607 1608 874 Bacoli are discussed in a later section because, in addition to anomalous FeOtot 1609 1610 875 concentrations, these MI show anomalous compositions for most of the other major 1611 876 elements. The lower FeO concentrations (from 4.1 to 6.2 wt%) of Fossa Lupara MI are 1612 1613 877 consistent with overheating during experiments owing to the low FeO concentrations of 1614 878 clinopyroxenes hosting these MI (3.7 to 4.5 wt%). These MI show elevated 1615 1616 879 concentrations of CaO and MgO and a general depletion of the other major elements 1617 1618 880 (Fig. 5 c and d). The effect of overheating can be observed in most of the MI hosted in 1619 881 clinopyroxene from this study and in MI from the literature. In fact, most of the reheated 1620 1621 1622 29 1623 1624 1625 1626 1627 882 MI (red triangles in Fig 5 d) that fall outside of the field defined by bulk rocks and can be 1628 1629 883 distinguished from the unheated MI (blue triangles in Fig. 5 d) based on their generally 1630 1631 884 lower FeOtot concentrations and slightly elevated CaO contents. 1632 885 Roedder (1979) reported that homogenization temperatures obtained from MI 1633 1634 886 might be higher than the temperature of trapping owing mainly to three factors: (1) 1635 887 inadequate time for re-equilibration (heating rates during experiment are too fast), (2) 1636 1637 888 loss of hydrogen or water from hydrous MI (which increases the liquidus temperature 1638 1639 889 relative to more H2O-rich compositions), and (3) thermal gradients in the stage. In 1640 890 addition to these factors, heating beyond the original trapping temperature would be 1641 1642 891 required to homogenize the MI if it did not originally trap a single homogeneous silicate 1643 892 melt phase (Klébesz et al., 2015; Student and Bodnar, 2004). Anomalously high 1644 1645 893 homogenization temperatures can also result if the MI original trapped a silicate melt plus 1646 894 a mineral inclusion (see Figure 1-15 in Bodnar and Student, 2006). Small minerals 1647 1648 895 trapped along with melt might be easily misidentified as daughter crystals, although this 1649 1650 896 interpretation can be ruled out by comparing the final quenched glass compositions 1651 897 because MI containing trapped solid(s) will be enriched in components that are major 1652 1653 898 constituents of the trapped mineral phase(s) (e.g., P2O5 enrichment if the co-trapped 1654 899 mineral inclusion is apatite). Finally, homogenization temperatures in excess of trapping 1655 1656 900 temperatures will result during re-heating if an MI has been affected by leakage after 1657 1658 901 trapping (see Figure 1-3, D and E in Bodnar and Student, 2006). This last possibility can 1659 902 often be ruled out with careful petrographic analysis of the studied MI. Even though it is 1660 1661 903 important to understand which factors could have affected the MI analyzed here to 1662 904 produce elevated homogenization temperatures, the overheating observation highlights 1663 1664 905 the importance of performing heating runs through kinetic experiments if the correct (or 1665 1666 906 approximate) temperature of trapping is required (Danyushevsky et al., 2002a). An 1667 907 additional consideration is that heating stage experiments are performed at atmospheric 1668 1669 908 pressure. Because the phenocryst host precipitated at pressures higher than ambient (1 1670 909 atm) pressure, the pressure in the MI at ambient laboratory pressure would necessarily be 1671 1672 910 lower than that at trapping owing to the lower confining pressure, thus requiring heating 1673 1674 911 to temperatures higher than the formation temperature to produce a single homogeneous 1675 912 melt phase. 1676 1677 1678 30 1679 1680 1681 1682 1683 913 In addition to overheating, post-entrapment loss of FeO from the MI could have 1684 1685 914 contributed to the observed depletion in FeOtot. In particular, a few MI, which were not 1686 1687 915 heated in the laboratory, show low FeOtot concentrations relative to the differentiation 1688 916 trend defined by bulk rocks, suggesting diffusive Fe loss. MI data from the literature also 1689 1690 917 suggest that diffusive Fe loss has affected some MI from CF and IP. Esposito et al. 1691 918 (2011) studied samples representative of the Solchiaro eruption and reported a negative 1692 1693 919 correlation between FeOtot concentrations of MI and the Fo content of the olivine host. 1694 1695 920 For this reason, Esposito et al. (2011) corrected MI compositions for diffusive Fe loss, 1696 921 and here we use the corrected compositions. Similarly, Mangiacapra et al. (2008) studied 1697 1698 922 representative samples of Fondo Riccio and Minopoli2 eruptions and reported a group of 1699 923 MI showing FeOtot < 1 wt%, which suggests some MI were affected by Fe loss. It is 1700 1701 924 important to note that the Fe-poor MI reported by Mangiacapra et al. (2008) do not show 1702 925 anomalous concentrations for the other major elements. In addition, MI data from the 1703

1704 926 literature show a wide range of FeOtot concentrations for SiO2 <55 wt%, and more 1705 1706 927 evolved MI show FeOtot concentrations that are consistent with the bulk rock FeOtot 1707 928 versus SiO2 trend, indicating that more evolved MI are less affected by Fe loss/gain. 1708 1709 929 Because MI examined here show evidence that the original MI composition has been 1710 930 modified either by diffusive Fe loss or overheating, we question whether volatile contents 1711 1712 931 obtained from the MI are representative of CF and IP pre-eruptive melts. The three 1713 1714 932 overheated MI from the Fossa Lupara eruption studied here show the lowest volatile 1715 933 contents among all the MI considered (including the MI data from the literature), 1716 1717 934 supporting the interpretation that Fe-poor MI and MI that were overheated in the 1718 935 laboratory should be avoided when studying volatile evolution in magmatic systems. 1719 1720 936 Considering data from this study and from the literature, not only do MI from CF 1721 1722 937 and IP show evidence for Fe loss, but also some MI show evidence that the Fe 1723 938 concentration has increased. For instance, a group of MI hosted in olivine from the 1724 1725 939 Solchiaro eruption (Mormone et al., 2011) shows higher FeOtot concentrations relative to 1726 940 the trend defined by bulk rock compositions (Fig. 5 b). These MI were not reheated in the 1727 1728 941 laboratory. It is important to note that the Fe-rich MI reported by Mormone et al. (2011) 1729 1730 942 are also depleted in MgO. In particular, the MgO concentrations of the MI reported by 1731 943 Mormone et al. (2011) range from 4.2 to 6.5 wt%, with three MI showing MgO from 0.5 1732 1733 1734 31 1735 1736 1737 1738 1739 944 to 1.7 wt% (see Table 1 by Mormone et al., 2011), values that are much lower than the 1740 1741 945 respective bulk rock compositions (8.58 wt%; sample Pro7/11 by De Astis et al., 2004). 1742 1743 946 Mormone et al. (2011) interpreted the MI based on EMP data, arguing that the correction 1744 947 for PEC was not necessary. Moreover, Fe-rich and Mg-poor MI described by Mormone 1745 1746 948 et al. (2011) are difficult to reconcile based on PEC because, when olivine crystallizes on 1747 949 the MI wall, both the MgO and the FeOtot contents of an MI should decrease because the 1748 1749 950 FeO content of the olivine host that crystallizes on the wall is generally higher than the 1750 1751 951 FeOtot content of the initially trapped melt. 1752 952 Different processes can be proposed to explain the enrichment in FeOtot of a MI. 1753 1754 953 First, the Fe-rich MI can result from Fe diffusing into the MI from the olivine host during 1755 954 re-equilibration due to natural overheating and melting of some olivine from the MI wall. 1756 1757 955 In fact, as more olivine component is added to the MI, the melt must decrease its Mg# to 1758 956 be in equilibrium with its host (Danyushevsky and Plechov, 2011 and references therein). 1759 1760 957 For the same reason, Fe gain will be accompanied by Mg loss. Second, natural 1761 1762 958 overheating can cause dissolution of a pre-existing Fe-oxide. In their study of re- 1763 959 homogenized MI hosted in olivine from scoria and basaltic lavas, Rowe et al. (2006) 1764 1765 960 reported that several MI show higher FeOtot contents relative to bulk rocks and suggested 1766 961 two explanations (1) dissolution of olivine host and (2) dissolution of co-trapped mineral 1767 1768 962 inclusions such as chromite, sulfide, and magnetite. In the case of the Fe-rich MI from CF 1769 1770 963 and IP, the Fe gain can be explained by natural overheating of MI originally containing 1771 964 co-trapped chromite. This natural overheating could be due to less evolved/hotter magma 1772 1773 965 mixing with the olivine-bearing magma or to olivine sinking to hotter levels of the 1774 966 magma reservoir. These processes are consistent with reverse zoning of olivine as 1775 1776 967 reported in several studies (e.g., D'Antonio and Di Girolamo, 1994) and the presence of 1777 1778 968 chromite inclusions in olivine from Solchiaro (Esposito et al., 2011). The Fe-rich MI 1779 969 from CF and IP show significantly lower SiO2 compared to other MI from these 1780 1781 970 locations, and the SiO2 vs Al2O3 trend points towards chromite composition, supporting 1782 971 an interpretation that melting of a co-trapped chromite is responsible for the elevated Fe 1783 1784 972 contents. It is also important to note that some of the Fe-rich MI were defined by 1785 1786 973 Mormone et al. (2011) as exhibiting "high" or "low" degrees of crystallization. An 1787 1788 1789 1790 32 1791 1792 1793 1794 1795 974 enrichment of FeO can result from measuring a heterogeneous (“high” or “low” 1796 tot 1797 975 crystallized MI) material containing Fe-oxides. 1798 1799 976 1800 977 TiO2 variability of less evolved MI 1801 1802 978 In this study, MI show TiO2 contents spanning a wide range, particularly for less 1803 979 evolved MI that vary from ~0.5 to ~1.7 wt% TiO2. This behavior is consistent with the 1804 1805 980 wide range in TiO2 concentrations of MI reported in other studies (Cannatelli et al., 2007; 1806 1807 981 Esposito et al., 2011; Mormone et al., 2011). In particular, Esposito et al. (2011) reported 1808 982 TiO2-rich MI containing up to ~2.4 wt% TiO2 and TiO2-poor MI with concentrations as 1809 1810 983 low as 0.4 wt%, and interpreted these anomalies to be the result of diffusion-mixing 1811 984 reaction in the mush zone. In contrast to MI, TiO2 concentrations of less evolved bulk 1812 1813 985 rocks at CF show less variation for the same degree of differentiation (Fig. 2 d). It is 1814 986 worth noting that TiO and SiO show positive trends for less evolved IP bulk rocks, 1815 2 2

1816 987 while the bulk rocks of CF show a quasi-horizontal TiO2 vs. SiO2 trend (see Fig. 2 d). 1817 1818 988 One interpretation is that this difference indicates that crystallization of Ti-magnetite 1819 989 started earlier in the evolution of CF magmas compared to magmas associated with IP. 1820 1821 990 1822 991 Anomalous plagioclase-hosted MI from the Bacoli eruption. 1823 1824 992 As reported in the results section, three MI hosted in plagioclase from the Bacoli 1825 1826 993 eruption do not adhere to most of the differentiation trends defined by bulk rock 1827 994 compositions (Fig. 2 a to e). The MI are Qz-normative, which is rare for melts of CF and 1828 1829 995 IP. The origin of these MI may be interpreted in three ways: (1) the melt may have been 1830 996 locally affected by dissolution-reaction-mixing (DRM) (Danyushevsky et al., 2004 and 1831 1832 997 references therein), (2) the plagioclase containing these MI may be xenocrystic and 1833 1834 998 formed in a different (temporal and/or spatial) part of the overall magmatic system, and 1835 999 (3) the melt in plagioclase may be representative of the magma chamber/country rock 1836 1837 1000 interface. It is important to note that these three MI show lower major oxide totals 1838 1001 relative to the other MI of this study (from 89.6 to 94.4 wt%). Unfortunately, we cannot 1839 1840 1002 assess the origin of these anomalous MI at this time. 1841 1842 1003 1843 1004 Summary of the major element anomalies observed in MI data 1844 1845 1846 33 1847 1848 1849 1850 1851 1005 In summary, the comparison between MI data and whole rock data indicates that 1852 1853 1006 some MI show anomalous compositions that do not follow the bulk rock trend, and these 1854 1855 1007 compositions cannot be explained by the main differentiation trend. The anomalous MI 1856 1008 compositions cannot be corrected by simply adding aliquots of mineral host back into the 1857 1858 1009 melt (PEC correction) or by subtracting the host mineral composition from the melt 1859 1010 composition (host dissolution correction). MI showing anomalous compositions either 1860 1861 1011 trapped a melt that was not in equilibrium with the host (DRM or boundary layer 1862 1863 1012 processes), or the composition was modified by processes other than simple PEC or host 1864 1013 dissolution after trapping. For these reasons, we conclude that anomalous MI likely 1865 1866 1014 record volatile contents that are not representative of the melt that was in equilibrium 1867 1015 with the growing phenocryst host. Moreover, most of the processes invoked to produce 1868 1869 1016 anomalous MI can also affect the volatile concentrations of melt (glass) owing to the 1870 1017 dependence of volatile solubility on the major element composition of melts (Papale et 1871 1872 1018 al., 2006; Shishkina et al., 2010). In addition, trapping of apatite inclusions within the MI 1873 1874 1019 as a component of DRM can affect the Cl, F, and H2O concentration of the melt. We 1875 1020 recommend that MI showing anomalous major element concentrations should not be 1876 1877 1021 selected for volatile analysis because the volatile concentrations of these MI are likely not 1878 1022 representative of the magmatic system. If the major element composition of the MI is not 1879 1880 1023 known, one cannot know that the MI composition is anomalous, and, thus whether the 1881 1882 1024 volatile content of that MI is representative. Therefore, it is fundamental that each MI is 1883 1025 analyzed to determine major element composition before it is analyzed for volatiles. Even 1884 1885 1026 though this study is focused on CF and IP volcanic fields, anomalous MI are reported in 1886 1027 many other volcanic systems (Danyushevsky et al., 2004 and references therein), and 1887 1888 1028 these same conclusions and recommendations apply to all such anomalous MI. 1889 1890 1029 In the following sections, we interpret and discuss the volatile budgets of MI from 1891 1030 this study and from the literature. We do not consider MI showing anomalous major 1892 1893 1031 element compositions in this assessment. Similarly, we exclude any MI from the 1894 1032 literature for which major element compositions are not reported. Additionally, MI 1895 1896 1033 hosted in plagioclase were not included because the three MI hosted in plagioclase that 1897 1898 1034 we analyzed all showed anomalous compositions relative to the bulk rocks. It is 1899 1035 important of note that our data filtering does not eliminate MI affected by PEC but only 1900 1901 1902 34 1903 1904 1905 1906 1907 1036 MI that could not be corrected by simply adding or subtracting the mineral host to the 1908 1909 1037 anomalous MI. In the following discussion we use ratios of elements that are 1910 1911 1038 incompatible with the host because these ratios should not be affected by PEC (e.g., Lima 1912 1039 et al., 2003). Also, as noted above, while it is possible to apply relatively robust PEC 1913 1914 1040 corrections for MI hosted in olivine (Danyushevsky and Plechov, 2011), reliable models 1915 1041 are not available to correct for PEC for MI hosted in clinopyroxene and feldspar. 1916 1917 1042 1918 1919 1043 Evolution of volatile concentrations of MI 1920 1921 1044 Several studies reported bubble-bearing MI in samples from CF and IP, but fewer 1922 1923 1045 studies have identified which MI (bubble-bearing or bubble-free) correspond to a specific 1924 1046 volatile analysis. One exception is the work of Esposito et al. (2011), who identified each 1925 1926 1047 MI as either bubble-bearing or bubble-free before exposing the MI at the surface of the 1927 1048 host for analysis. Based on recent studies (e.g., Aster et al., 2016; Moore et al., 2015), it 1928 1929 1049 is not possible to determine if the reported volatile content of an MI represents the actual 1930 1931 1050 volatile content of the trapped melt or represents only a portion of the total volatile 1932 1051 budget in the original melt if the type of MI (bubble-bearing versus bubble-free) is not 1933 1934 1052 reported. Thus, the following discussion relies only on data from bubble-free MI that do 1935 1053 not show anomalous major element compositions. 1936 1937 1054 By including only bubble-free MI data in this interpretation, it is not necessary to 1938 1939 1055 account for any volatiles that might have been lost to the bubble after trapping of a single 1940 1056 homogeneous silicate melt phase. Even though bubble-free MI may have been affected 1941 1942 1057 by H loss, we argue that extensive diffusive H loss would lead to the formation of a vapor 1943 1058 bubble (Steele-MacInnis et al., 2017). Thus, the lack of a bubble supports the 1944 1945 1059 interpretation that H loss was not significant for the bubble-free MI considered here. 1946 1947 1060 Moreover, we do not observe any correlation between CO2 and H2O as would be 1948 1061 expected if all MI started with the same volatile contents, and each MI experienced 1949 1950 1062 different degrees of H loss (Bucholz et al., 2013). Also, If diffusive H loss is only 1951 1063 controlled by redox reactions of the MI-host system (Danyushevsky et al., 2002a), then it 1952 1953 1064 is unlikely that a significant amount of H2O was lost from the MI because magnetite that 1954 1065 would be produced by this process it is not observed in the studied MI. 1955 1956 1957 1958 35 1959 1960 1961 1962 1963 1066 Finally, it is important to note that filtering the MI data based on the presence or 1964 1965 1067 absence of a bubble may mask the complete variability represented by all the MI. In fact, 1966 1967 1068 it is more likely that a volatile-rich MI will form a bubble after trapping, compared to an 1968 1069 MI that is relatively volatile-poor. Considering only bubble-free MI may unintentionally 1969 1970 1070 exclude some MI that trapped melts with higher volatile contents. We suggest that the 1971 1071 best practice would be to analyze the volatile contents of the MI bubble and correct the 1972 1973 1072 original volatile content using methods described in the literature (e.g., Esposito et al., 1974 1975 1073 2011; Mironov et al., 2015; Moore et al., 2015). In the next section, we interpret the H2O- 1976 1074 CO2 data of MI from CF and IP volcanic fields considering only bubble-free MI. 1977 1978 1075 1979 1076 H2O and CO2 degassing, crystallization and mixing of basaltic-trachybasaltic 1980 1981 1077 magmas 1982 1078 The volatile content of the glass phase in MI can be used as a geobarometer 1983

1984 1079 because CO2 and H2O solubilities in silicate melts are highly pressure dependent (Blundy 1985 1986 1080 and Cashman, 2008). A large range in CO2 contents of MI, combined with a narrow 1987 1081 range of H2O concentrations, is often interpreted to reflect MI trapping along a degassing 1988 1989 1082 (ascent) path because CO2 is lost from the melt early and before significant H2O loss 1990 1083 occurs. In this scenario, MI are trapped during crystallization of phenocrysts at different 1991 1992 1084 depths (pressures) as the magma ascends through the crust. Alternatively, different 1993 1994 1085 phenocrysts might trap MI from essentially the same silicate melt at various depths 1995 1086 within a stagnant plumbing system. Thus, H2O-CO2 systematics within a single eruptive 1996 1997 1087 event can support or dismiss the hypothesis that crystallization in the plumbing system 1998 1088 occurred under volatile-saturated conditions, and if the crystallization occurs over a large 1999 2000 1089 or narrow range of pressure (depths.) 2001 2002 1090 For CF and IP, MI data from individual eruptions are often interpreted to 2003 1091 represent trapping during ascent as the magma degasses (Arienzo et al., 2016; Arienzo et 2004 2005 1092 al., 2010; Cecchetti et al., 2001; Cipriani et al., 2008; Esposito et al., 2011; Esposito et 2006 1093 al., 2014; Mangiacapra et al., 2008; Moretti et al., 2013; Mormone et al., 2011) and that 2007 2008 1094 crystallization occurred under volatile-saturated conditions, and over a large range of 2009 2010 1095 pressures. As reported above, earlier studies interpreted the variability in H2O vs. CO2 2011 1096 systematics observed in MI as a combination of trapping at various depths along a 2012 2013 2014 36 2015 2016 2017 2018 2019 1097 degassing path, combined with episodic CO fluxing events (Arienzo et al., 2016; 2020 2 2021 1098 Arienzo et al., 2010; Mangiacapra et al., 2008; Moretti et al., 2013; Mormone et al., 2022 2023 1099 2011). However, we emphasize that “shrinkage” bubbles are often reported in MI from 2024 1100 studies advocating for CO2 fluxing, and most workers do not distinguish between bubble- 2025 2026 1101 free and bubble-bearing MI, nor do they account for volatiles contained in the bubble. 2027 1102 Because these studies do not discuss these issues, their interpretations concerning 2028 2029 1103 degassing and CO2 fluxing cannot be tested because the true, pre-eruptive volatile 2030 2031 1104 contents of the melt are unknown. In fact, Esposito et al. (2011), Hartley et al. (2014), 2032 1105 Moore et al. (2015) and Esposito et al. (2016) all demonstrated that the bubble may 2033 2034 1106 contain as much as 99% of the originally trapped CO2. For this reason, the discussion that 2035 1107 follows is based on data from bubble-free MI only. 2036 2037 1108 When we consider bubble-free MI of basaltic and trachybasaltic composition, the 2038 1109 distribution of H O-CO data shows less variability relative to data from all types of MI 2039 2 2

2040 1110 (bubble-free and bubble-bearing). The maximum CO2 content of bubble-free MI (1248 2041 2042 1111 ppm) is lower than that of bubble-bearing MI (1706 ppm). Also, the H2O content of 2043 1112 bubble-free MI shows less variability relative to the bubble-bearing group. Thus, the 2044 2045 1113 large variability in CO2 contents combined with the narrow range of H2O contents of 2046 1114 bubble-free MI is more pronounced relative to bubble-bearing MI, suggesting that the 2047 2048 1115 magma was volatile-saturated and that olivine crystallized at various depths. For this 2049 2050 1116 discussion, we consider bubble-free MI as well as one bubble-bearing MI for which the 2051 1117 original CO2 content was corrected based on Raman analysis of the vapor bubble (MI 2052 2053 1118 RESC2-O26-MI3; Esposito et al., 2011). 2054 1119 In order to test the hypothesis that trapping of MI occurs during fractional 2055 2056 1120 crystallization under volatile-saturated conditions at various sub-volcanic depths, we 2057 2058 1121 computed fractional crystallization paths using the rhyolite-MELTS code (Ghiorso and 2059 1122 Gualda, 2015). We considered MgO-rich bulk rocks from the IP and noted that the 2060 2061 1123 Al2O3/K2O ratios of these rocks span a wide range (Figs. 6 and 7). We performed 2062 1124 fractional crystallization calculations using rhyolite-MELTS to test if the general trend in 2063 2064 1125 Al2O3/K2O vs. MgO systematics of bulk rocks is consistent with trends predicted during 2065 2066 1126 fractional crystallization. The first question we addressed was whether the previously 2067 1127 reported MgO-rich bulk rock compositions can be assumed to represent liquid 2068 2069 2070 37 2071 2072 2073 2074 2075 1128 compositions. D’Antonio and Di Girolamo (1994) and D'Antonio et al. (1999a) noted 2076 2077 1129 that, for some of the bulk rocks, olivine phenocrysts may represent cumulus material and 2078 2079 1130 would thus be considered xenocrysts. In the studies by D’Antonio and Di Girolamo 2080 1131 (1994) and D'Antonio et al. (1999a), the olivine showing the highest forsterite content 2081 2082 1132 was compared to the olivine composition predicted using various partition coefficients, 2083 1133 KDFe-Mg, following the method of Rhodes (1981). These calculations indicated that 2084 2085 1134 olivine compositions in some of the bulk rocks were sometimes significantly different 2086 2087 1135 from the predicted olivine compositions, but the forsterite content of olivine never 2088 1136 exceeded the calculated values (Table 3). Using the same bulk rock data from the 2089 2090 1137 literature, we compared the olivine composition calculated at the liquidus using rhyolite- 2091 1138 MELTS with the observed data, as reported in Table 3. We noted that the olivine 2092 2093 1139 compositions of the bulk rocks are comparable to the olivine composition predicted by 2094 1140 rhyolite-MELTS, assuming 200 MPa, NNO, and H O-CO saturation (see discussion 2095 2 2

2096 1141 below). In particular, bulk rock sample Apr22 shows Fomax of 86.6. The rhyolite-MELTS 2097 2098 1142 predictions showed that a liquid composition equivalent to Apr22 is in equilibrium with 2099 1143 Fo87.3 at its liquidus (Table 3). D’Antonio and Di Girolamo (1994) reported that bulk rock 2100 2101 1144 Apr22 contains 11 vol% phenocrysts (olivine plus clinopyroxene). In other samples 2102 1145 olivine ranges from 2.2 to 12.8 vol%, and clinopyroxene ranges from 2.5 to 7.6 vol%. We 2103 2104 1146 used the rhyolite-MELTS predictions as a starting composition to evaluate the 2105 2106 1147 consistency of predicted volume proportions of olivine and clinopyroxene and compared 2107 1148 these values to values reported in the literature. According to these calculations, the 2108 2109 1149 magma should contain 12 vol% phenocrysts (olivine plus clinopyroxene) and 88 vol% 2110 1150 liquid after ~115°C cooling. At this stage, the magma contains ~4 vol% olivine and ~8 2111 2112 1151 vol% clinopyroxene. The liquid would initially be in equilibrium with an Fo87 olivine and 2113 2114 1152 the final olivine to crystallize would be Fo84. These results are consistent with the 2115 1153 information reported in the literature and support the interpretation that trapping of MI 2116 2117 1154 occurs during fractional crystallization. 2118 1155 We selected five starting compositions for rhyolite-MELTS calculations: three 2119 2120 1156 bulk rock compositions, and two MI compositions that were corrected for PEC. Sample 2121 2122 1157 Apr18 (D’Antonio and Di Girolamo, 1994) was selected because this bulk rock 2123 1158 composition shows both the highest MgO content and highest Al2O3/K2O ratio. We also 2124 2125 2126 38 2127 2128 2129 2130 2131 1159 selected Apr22 (D’Antonio and Di Girolamo, 1994) because this rock showed the best 2132 2133 1160 match between the maximum forsterite content of olivine and the composition of olivine 2134 2135 1161 calculated to be in equilibrium with the rock composition on its liquidus (Table 3). In 2136 1162 addition, we selected an initial composition that was based on the MI exhibiting the 2137 2138 1163 highest Al2O3/K2O ratio [RESC5-O5-MA; Esposito et al. (2011)]. It is important to note 2139 1164 that while bulk rock compositions can be affected by secondary processes (such as post 2140 2141 1165 depositional water-rock interactions) or entrainment of xenocrysts, the compositions of 2142 2143 1166 MI should not be affected by entrainment of xenocrysts into the surrounding liquid. As 2144 1167 described by Fowler et al. (2007), we started with the uncorrected composition of MI 2145 2146 1168 RESC5-O5-MA (Esposito et al., 2011) and added aliquots of olivine host until the MI 2147 1169 composition was in equilibrium with the host olivine measured adjacent to the MI. This 2148 2149 1170 allowed us to correct the MI composition for the effects of PEC. The fourth composition 2150 1171 selected was bulk-rock sample Ps3 reported by Di Girolamo et al. (1984) because this 2151

2152 1172 composition shows relatively high MgO and relatively low Al2O3/K2O. Finally, we 2153 2154 1173 selected the uncorrected composition of MI RESC3-O9-MA described by Esposito et al. 2155 1174 (2011). This MI showed relatively low Al2O3/K2O and the host olivine had a high 2156 2157 1175 forsterite content. We applied the same correction for PEC that was applied to MI 2158 1176 RESC5-O5-MA. 2159 2160 1177 Using these five initial compositions, we ran the models using specific starting 2161 2162 1178 conditions. For the bulk rocks (Apr18, Apr22 and Ps3), we calculated the liquid line of 2163 1179 descent using rhyolite-MELTS, assuming H2O (1.25 wt%) and CO2 (1400 ppm) contents 2164 2165 1180 based on the H2O-CO2 systematics of bubble-free MI hosted in olivine. We assumed 200 2166 1181 MPa as the initial pressure, consistent with solubility experiments under volatile- 2167 2168 1182 saturated conditions for a trachybasaltic melt at the H2O-CO2 concentrations of our 2169 2170 1183 models (Fanara et al., 2015). As shown in Fig. 8, isobaric crystallization in the absence of 2171 1184 volatile saturation produces trends that are inconsistent with the mafic MI data. 2172 2173 1185 Specifically, during crystallization at volatile-undersaturated conditions, both H2O and 2174 1186 CO concentrations increase simultaneously. Conversely, the calculated trends based on 2175 2 2176 1187 bubble-free, basaltic-trachybasaltic MI data are consistent with polybaric crystallization 2177 2178 1188 under volatile-saturated conditions. For the model using the MI composition as the 2179 1189 starting composition (PEC-corrected MIs RESC5-O5-MA and RESC3-O9-MA), we 2180 2181 2182 39 2183 2184 2185 2186 2187 1190 normalized the measured volatile contents to 100%, resulting in 1.21 wt% H O and 954 2188 2 2189 1191 ppm CO2 for RESC5-O5-MA, and 1.13 wt% H2O and 479 ppm CO2 for RESC3-O9-MA, 2190 2191 1192 respectively. The pressure at volatile saturation (thus, the initial pressure in the models) 2192 1193 was 160 MPa for RESC5-O5-MA and 90 MPa for RESC3-O9-MA. 2193 2194 1194 Effect of starting composition. The first step was to test if the three high- 2195 1195 Al2O3/K2O starting compositions (Apr18, Apr22, and RESC5-O5-MA) could reproduce 2196 2197 1196 the geochemical trends defined by bulk rocks, the phases that crystallized, and the 2198 2199 1197 compositions of the phases. For this test, we assumed isobaric crystallization and a fixed 2200 1198 oxidation state. The results are shown in Figs. 6, 9-11. The three starting compositions do 2201 2202 1199 not reproduce consistent geochemical trends for MgO and SiO2/CaO vs. Al2O3/K2O (Fig. 2203 1200 6a and d). The range of olivine compositions is consistent with the range from previous 2204 2205 1201 studies (Fig. 12). However, the calculated trend in Fo mol% vs. Al2O3/K2O space does 2206 1202 not agree with data from the bubble-free MI (Fig. 12). The clinopyroxene compositions 2207 2208 1203 predicted by rhyolite-MELTS show a good match with measured compositions (Fig. 9a), 2209 2210 1204 but the calculated feldspar compositions do not reproduce the observed alkali-feldspar 2211 1205 composition (Fig. 10a). On the TAS diagram, only data from sample Apr22 produces a 2212 2213 1206 trend that is consistent with the “normal” MI compositions from the literature and from 2214 1207 this study, but all three starting compositions reproduce the trachybasalt-shoshonite- 2215 2216 1208 latite-phonolite/trachyte bulk rock trend (Fig. 11a). It is important to note that the MI data 2217 2218 1209 plotted in the TAS diagram can be used to track the stage in the differentiation sequence 2219 1210 that the host phase for each MI crystallized, and we compared the relative timing of 2220 2221 1211 olivine, clinopyroxene and sanidine formation predicted by rhyolite-MELTS with results 2222 1212 based on MI. For the Apr22 experiment, sanidine crystallization was predicted only 2223 2224 1213 during the last stages of crystallization, which does not match with compositions of MI 2225 2226 1214 hosted in sanidine. In addition, rhyolite-MELTS predicts that clinopyroxene 2227 1215 crystallization ended earlier than sanidine crystallization, and that these two phases did 2228 2229 1216 not crystallize in equilibrium, in disagreement with petrographic analysis of more 2230 1217 evolved rocks (e.g., Di Girolamo et al. 1984). Trends predicted by rhyolite-MELTS for 2231 2232 1218 MI and Apr18 starting compositions do not reproduce the observed crystallization trend 2233 2234 1219 of alkali-feldspar. 2235 2236 2237 2238 40 2239 2240 2241 2242 2243 1220 We also examined if the low-Al O /K O bulk rock composition and/or MI 2244 2 3 2 2245 1221 compositions would produce a better match between rhyolite-MELTS predictions and 2246 2247 1222 data from natural samples. For MgO and SiO2/CaO vs. Al2O3/K2O, the predicted 2248 1223 fractional crystallization trends are consistent for the general bulk rock data for 2249 2250 1224 Al2O3/K2O < 7 (Fig. 13). As shown in Figs. 7a and d and 13a, the match applies not only 2251 1225 to bulk rock data, but also to MI data. What is even more striking is the match between 2252 2253 1226 rhyolite-MELTS predictions and “normal” MI in the TAS diagram. Experiments 2254 2255 1227 assuming Ps3 and RESC3-O9-MA as the starting composition better predicted the 2256 1228 occurrence of sanidine crystallization, relative to calculations based on Apr22 and Apr18 2257 2258 1229 and RESC5-O5-MA reported above. In addition, RESC3-O9-MA experiments predicted 2259 1230 clinopyroxene crystallization throughout the fractionation trend and that clinopyroxene 2260 2261 1231 coexisted with sanidine. The only mismatch for this run is that olivine crystallization 2262 1232 ended later in the differentiation sequence than is observed in the samples. However, as 2263 2264 1233 shown in the supplementary material (Appendix Table C), the mass fraction of olivine 2265 2266 1234 crystallizing as the path moves through the shoshonite field continuously decreases. 2267 1235 Thus, predictions based on Ps3 and RESC3-O9-MA starting compositions are in better 2268 2269 1236 agreement with observations, compared to Apr22, Apr18, and RESC5-O5-MA for both 2270 1237 the TAS trend and mineral phase appearance (Fig. 11d). The same conclusion can be 2271 2272 1238 drawn for the feldspar compositions (Fig. 10d). The plagioclase compositions predicted 2273 2274 1239 by Ps3 and RESC3-O9-MA starting compositions are more representative of plagioclase 2275 1240 compositions reported in the literature. In addition, the trends extend toward a higher 2276 2277 1241 sanidine component and show better agreement with sanidine compositions from the 2278 1242 literature. The trend predicted for Fo mol% vs. Al O /K O, assuming Ps3 and RESC3- 2279 2 3 2 2280 1243 O9-MA starting compositions, is consistent with several of the MI compositions (Fig. 2281 2282 1244 12). 2283 1245 Effect of pressure (depth) evolution during crystallization. In order to better 2284 2285 1246 understand if crystallization occurred while the magma was ponded in the subsurface (at 2286 1247 a constant pressure) or occurred in the plumbing system as the magma ascended through 2287 2288 1248 the crust, we ran rhyolite-MELTS experiments assuming both isobaric and polybaric 2289 2290 1249 trends. We first selected sample Apr18 and assumed isobaric crystallization under 2291 1250 volatile-saturated conditions at 200 MPa. In addition, we ran the same model under 2292 2293 2294 41 2295 2296 2297 2298 2299 1251 polybaric conditions with crystallization occurring as the pressure dropped from 200 to 2300 2301 1252 14 MPa. The oxidation state was fixed by the NNO buffer in both examples. We note that 2302 2303 1253 the polybaric model improves the fit for alkali feldspar compositions (Fig. 10b). 2304 1254 However, the TAS trend does not improve, and sanidine crystallization is predicted to 2305 2306 1255 occur later than is suggested by MI hosted in sanidine (Fig. 11b). In addition, the 2307 1256 predicted SiO2 concentration extends toward rhyolitic liquid. It is worth noting that for 2308 2309 1257 polybaric conditions, rhyolite-MELTS does not predict the formation of Ca-Ti-rich 2310 2311 1258 clinopyroxene. 2312 1259 We also considered Ps3 as the starting composition and compared the isobaric 2313 2314 1260 trend at 200 MPa with two polybaric trends. The polybaric model conditions were 2315 1261 defined based on H2O vs. CO2 systematics (Fig. 8). The first polybaric trend assumed a 2316 2317 1262 0.5 °C/MPa path from 200 MPa to 5 MPa, at which point the CO2 content of the melt was 2318 1263 2 ppm. After the pressure reached 5 MPa, continued fractional crystallization was 2319 2320 1264 assumed to be isobaric. The second polybaric model assumed a 0.1 °C/MPa path from 2321 2322 1265 200 MPa to 11 MPa, or until the CO2 concentration in the melt was 9 ppm. Continued 2323 1266 crystallization was assumed to be isobaric at 11 MPa. As can be seen in the SiO2/CaO vs. 2324 2325 1267 Al2O3/K2O trends, the polybaric models provide a better fit to the bulk rock and MI data 2326 1268 (Fig. 7b and e). The difference between isobaric and polybaric conditions is not 2327 2328 1269 significant for the TAS trends. However, differences are observed during the final 2329 2330 1270 differentiation steps, and sanidine appearance occurs at the limit of the field defined by 2331 1271 MI hosted in sanidine. The polybaric model shows a better fit to the sanidine 2332 2333 1272 compositions from the literature (Fig. 10e). Importantly, Ca-Ti-rich clinopyroxene is not 2334 1273 predicted in two of the polybaric crystallization models (Fig. 9e). 2335 2336 1274 Effect of oxidation state on predicted trends. The final test using rhyolite-MELTS 2337 2338 1275 examined potential effects of oxidation state on the differentiation trends by varying 2339 1276 oxygen fugacity over the range from QFM-2 and QFM+2. We assumed isobaric 2340 2341 1277 crystallization and Apr22 as the starting composition. We recognize that oxidation state is 2342 1278 expected to vary during extended periods of crystal fractionation (e.g., Moussallam et al., 2343 2344 1279 2014), but we have not conducted experiments that allow the oxygen fugacity to vary 2345 2346 1280 during a crystallization sequence. 2347 2348 2349 2350 42 2351 2352 2353 2354 2355 1281 Our results for Apr22 suggest that oxidation state does not significantly affect the 2356 2357 1282 various geochemical trends predicted by rhyolite-MELTS that were tested here. Trends 2358 2359 1283 for MgO and SiO2/CaO vs. Al2O3/K2O show similar patterns that do not match the bulk 2360 1284 rock trends (Fig. 6c and f). The various oxidation states tested did not reproduce TAS 2361 2362 1285 trends during crystallization of clinopyroxene for the most evolved compositions (Fig. 2363 1286 11c and f), and feldspar compositions predicted by the models do not improve the match 2364 2365 1287 with the real data (Fig. 10c). 2366 2367 1288 We also selected sample Ps3 (low Al2O3/K2O ratio) as the starting composition 2368 1289 with a range of oxidation states from QFM-2 and QFM+2. The different oxidation states 2369 2370 1290 have little effect on the liquid trend during the initial differentiation step (Fig. 7c and f). 2371 1291 The match between predicted feldspar compositions and feldspar compositions from the 2372 2373 1292 literature does not improve (Fig. 10f). Varying the oxidation state did affect the trajectory 2374 1293 of the liquid on the TAS diagram and the trends for different oxidation states delimit a 2375 2376 1294 field consistent with the MI data (Fig. 11f), an observation that is consistent with an 2377 2378 1295 evolving oxidation state during fractional crystallization. 2379 1296 Summary of rhyolite-MELTS modeling results. Selecting initial compositions 2380 2381 1297 based on bulk rocks and MI with high MgO concentrations and Al2O3/K2O ratios does 2382 1298 not yield trends consistent with the analytical data. In contrast, it is reasonable to infer 2383 2384 1299 that samples Ps3 (bulk rock) and RESC3-O9-MA (MI) represent mafic liquids that could 2385 2386 1300 produce geochemical trends toward trachytic and phonolitic melts during fractional 2387 1301 crystallization. The consistency between bulk rock data, data from bubble-free MI, and 2388 2389 1302 rhyolite-MELTS predictions supports this hypothesis. In addition, the phases predicted to 2390 1303 form during crystallization, and their compositions, are consistent with the literature data. 2391 2392 1304 Furthermore, the TAS trends based on “normal” MI and the correlation between MI 2393 2394 1305 composition and host mineral match with rhyolite-MELTS predictions. In particular, the 2395 1306 predicted order of appearance of mineral phases is in good agreement with the phase 2396 2397 1307 fields defined by the MI compositions and the compositions of the MI hosts. The best 2398 1308 agreement between the model results and the analytical data is obtained when the 2399 2400 1309 crystallization path is polybaric, especially for sanidine compositions and the predicted 2401 2402 1310 absence of Ca-Ti-rich clinopyroxene. The oxidation state does not appear to significantly 2403 1311 influence the results. 2404 2405 2406 43 2407 2408 2409 2410 2411 1312 To compare bubble-free MI data with rhyolite-MELTS trends and bulk rock data, 2412 2413 1313 we compared ratios of olivine-incompatible elements (e.g., SiO2/CaO and Al2O3/K2O) 2414 2415 1314 rather than using absolute concentrations of elements in MI (Fig. 13a and b). Comparing 2416 1315 olivine-incompatible element ratios minimizes the effect of PEC (Lima et al., 2003) and 2417 2418 1316 these ratios are independent of methods used to correct the MI compositions (e.g., 2419 1317 Danyushevsky et al., 2002a). As shown by the SiO2/CaO vs. Al2O3/K2O trends in Fig. 13, 2420 2421 1318 the major element ratios suggest that bubble-free MI cannot be explained solely by a 2422 2423 1319 single fractional crystallization trend starting from a single primitive melt. As for bulk 2424 1320 rocks, a group of MI (hereafter referred to as Mixed-MI) forms a trend consistent with a 2425 2426 1321 mixing model (Fig. 14), and another group of bubble-free MI (hereafter referred to as 2427 1322 Melt-1 MI) is consistent with simple fractional crystallization from Melt-1. These groups 2428 2429 1323 of bubble-free MI overlap with the same two groups of bulk rocks that can be identified 2430 1324 based on geochemistry (Fig. 13a). It is important to note that the overlap in bulk rock 2431 2432 1325 composition and compositions of several MI indicates that the examined bulk rocks can 2433 2434 1326 be assumed to be representative liquids of magmas associated with volcanism at CF and 2435 1327 PI. Consequentially, SiO2/CaO or MgO vs. Al2O3/K2O trends of bulk rocks cannot be the 2436 2437 1328 result of plagioclase crystal assimilation. Some of the trace element ratios of bubble-free 2438 1329 MI show the same behavior as ratios of major elements, showing two distinct trends that 2439 2440 1330 cannot be produced by simple crystallization from a single starting composition 2441 2442 1331 represented by either Melt-1 or Mixed-MI and bulk rocks showing high-Al2O3/K2O. For 2443 1332 instance, Rb, Pb, Ba, U, Th, Nb, Ce, La, Pr, and Ta show the same trend as K2O in the 2444 2445 1333 SiO2/CaO vs. Al2O3/K2O diagram. The Mixed-MI form a different trend relative to Melt- 2446 1334 1 MI (Fig. 14a). The same behavior is observed when trace element ratios are plotted 2447 2448 1335 against each other (e.g., Ba/Sr vs. Yb/Rb and Eu/Rb Fig. 14 b and c). We used a simple 2449 2450 1336 binary mixing model based on Albarède (1996) assuming MI compositions RESC5-O5- 2451 1337 MA (for Mixed-MI) and RESC3-O9-MA (for Melt-1) as end members. 2452 2453 1338 Spider diagrams also highlight the difference between Mixed-MI and Melt-1 MI 2454 1339 (Fig. 15). For instance, the Primitive Mantle-normalized spider diagram (Sun, 1980) 2455 2456 1340 shows that the normalized LILE are generally lower for Mixed-MI, but that the same 2457 2458 1341 Mixed-MI show generally higher normalized HFSE (Fig. 15 c). Comparing the less 2459 1342 evolved Melt-1 MI (RESC3-O9-MA) with the mixed MI that shows the strongest 2460 2461 2462 44 2463 2464 2465 2466 2467 1343 evidence of mixing (RESC5-O5-MA) highlights the difference in trace element contents, 2468 2469 1344 such as for Rb, P, Eu, and Yb (Fig. 15c). These two MI are hosted in olivine with similar 2470 2471 1345 forsterite content and, therefore, crystal fractionation cannot be invoked to explain the 2472 1346 difference in the trace element patterns. Similar observations are suggested from the REE 2473 2474 1347 spider diagrams (Sun and McDonough, 1989), whereby Mixed-MI show lower 2475 1348 normalized LREE but generally higher normalized HREE, relative to Melt-1 MI (Fig. 2476 2477 1349 15b). It is important to note that for intermediate REE, the two groups of MI overlap, 2478 2479 1350 with more erratic behavior shown by Mixed-MI. These observations indicate that the 2480 1351 difference between these two groups cannot be explained by simple fractional 2481 2482 1352 crystallization but, rather, are explained by mixing of two different primitive melts. This 2483 1353 is in agreement with D’Antonio et al. (2007), who suggested that primitive magmas of 2484 2485 1354 differing composition can be generated by contamination of the mantle source by various 2486 1355 proportions of slab-derived fluids or by sediment-derived melts. 2487 2488 1356 The systematics of Cl and S are also consistent with the mixing model proposed 2489 2490 1357 in this study. Here again, the Mixed-MI form well-defined trends and diverge from the 2491 1358 general trend determined by Melt-1 MI (Fig. 16). This behavior can be observed in the 2492 2493 1359 Cl/Yb vs. Al2O3/K2O plots, where the Mixed-MI show lower Cl/Yb and higher 2494 1360 Al O /K O compared to Melt 1 MI (Fig. 16a). This suggests that, not only is Mixed-MI 2495 2 3 2 2496 1361 different from Melt-1 in terms of major and trace element trends, but also differs in terms 2497 2498 1362 of volatile contents. 2499 1363 To further test the hypothesis that mafic magmas beneath IP crystallize under 2500 2501 1364 volatile-saturated conditions and over a range of pressures, we first compared the volatile 2502 1365 ratios (e.g., H O/CO ) of bubble-free MI with the composition of the olivine hosts (Fig. 2503 2 2 2504 1366 17a). It is important to note that the composition of the host was measured ~40 µm away 2505 2506 1367 from the MI/host interface (Esposito et al., 2011). No correlation is observed between 2507 1368 H2O/CO2 ratio and the mol% Fo of the host. 2508 2509 1369 The H2O/CO2 versus mol% Fo systematics could be explained by assuming 2510 1370 different pressure-temperature evolution paths (Fig. 17a). A possible explanation of the 2511 2512 1371 different pressure-temperature paths is that feeding conduits of different geometries (e.g., 2513 2514 1372 thickness) could transport magmas of different temperatures to the same depth (Spera, 2515 1373 1980). Thus, cogenetic magmas of slightly different temperatures may occur at the same 2516 2517 2518 45 2519 2520 2521 2522 2523 1374 depth below a volcano. This scenario is consistent with the “mush column” model 2524 2525 1375 proposed by Marsh (1996). As summarized by Putirka (2017; see his Fig.1a and c) and 2526 2527 1376 by Cashman et al. (2017), the plumbing system beneath a volcano represents a mosaic of 2528 1377 scattered bodies which may or may not be connected to each other. The most likely result 2529 2530 1378 is that co-genetic magma pools in a plumbing system may follow different pressure- 2531 1379 temperature paths, consistent with the variability of volatiles found in MI hosted in 2532 2533 1380 olivine that show the same forsterite contents (Fig. 17a). This scenario is also consistent 2534 2535 1381 with the presence of anomalous MI in olivine as discussed by Danyushevsky et al. 2536 1382 (2004). 2537 2538 1383 Volatile ratios were plotted versus Al2O3/K2O to test for consistency between 2539 1384 volatile content of MI and major element compositions of MI. The distribution of bubble- 2540 2541 1385 free MI data in the H2O/CO2 versus Al2O3/K2O diagram is plotted along the same 2542 1386 polybaric crystallization trends. As was observed for major, trace and other volatile 2543

2544 1387 element ratios, H2O/CO2 ratios of Melt-1 MI are consistent with polybaric crystallization 2545 2546 1388 trends of Melt-1 and Mixed-MI, which define a mixing trend (Fig. 17b). 2547 1389 We further investigated the chronology of the magmatic processes to assess 2548 2549 1390 whether crystal fractionation starting from Melt-1 occurred before or after mixing by 2550 1391 examining textural and petrographic relationships of olivine and MI. First, we note that 2551 2552 1392 reverse zoning of olivine phenocrysts that were erupted in the IP is reported by many 2553 2554 1393 researchers (e.g., D'Antonio and Di Girolamo, 1994). Secondly, we re-evaluated data 2555 1394 reported by Esposito et al. (2011) to search for correlations between type of zonation of 2556 2557 1395 the host olivine crystals of the bubble-free MI reported in their Table 3 and the type of 2558 1396 MI defined here. Also, we investigated the correlation between the position of the 2559 2560 1397 bubble-free MI within the host and the type of MI (Mixed-MI and Melt-1 MI) as defined 2561 2562 1398 here. This effort showed that 73% of the Mixed-MI are hosted in zoned olivine crystals 2563 1399 and half of the Melt-1 MI are hosted in unzoned olivine phenocrysts. No Mixed-MI are 2564 2565 1400 hosted in the cores of olivine phenocrysts while the majority of Melt-1 MI are, indeed, 2566 1401 hosted in olivine cores (Table 4). This textural and petrographic information suggests that 2567 2568 1402 fractional crystallization of Melt-1 began in the various magma pools below the volcano 2569 2570 1403 before the mixing event associated with the later Solchiaro eruption. 2571 2572 2573 2574 46 2575 2576 2577 2578 2579 1404 In conclusion, taken together, the volatile, major and trace element compositions 2580 2581 1405 of bubble-free MI, host-mineral compositions, bulk rocks, and trends predicted by 2582 2583 1406 rhyolite-MELTS models all suggest that crystallization of melt beneath Solchiaro first 2584 1407 occurred by simple fractional crystallization under volatile-saturated conditions at depths 2585 2586 1408 between ~1 and 5 km (Fig. 18a). This interpretation is consistent with other MI studies 2587 1409 that suggest that differentiation of magmas at shallow depths beneath CF and IP, as well 2588 2589 1410 as in the nearby Ischia Islands, is characterized by H2O-CO2 volatile saturation (Arienzo 2590 2591 1411 et al., 2016; Arienzo et al., 2010; Cannatelli et al., 2007; Cecchetti et al., 2002-2003; 2592 1412 Cipriani et al., 2008; Esposito et al., 2011; Fourmentraux et al., 2012; Mangiacapra et al., 2593 2594 1413 2008; Mormone et al., 2011; Roach, 2005). Crystallization continued under polybaric 2595 1414 conditions during magma ascent to shallower crustal levels. Well-constrained MI indicate 2596 2597 1415 that magmas in shallower reservoirs were slightly more evolved than those in deeper 2598 1416 reservoirs, as predicted by Cashman et al. (2017). After this first stage dominated by 2599

2600 1417 magma ascent and crystallization, recharging by a second high-Al2O3/K2O basaltic melt 2601 2602 1418 occurred at depths between ~1 to ~6 km (Fig. 18b). During this pre-eruptive stage, Melt-1 2603 1419 that had undergone varying degrees of crystallization mixed with various proportions of 2604 2605 1420 high-Al2O3/K2O basaltic melt. Although mixing has been suggested previously based on 2606 1421 the isotopic composition of bulk rocks from different localities and different degrees of 2607 2608 1422 differentiation (e.g., D’Antonio et al., 2007; D’Antonio and Di Girolamo 1994), our 2609 2610 1423 interpretation of mixing is based on MI and is representative of a single eruption (single 2611 1424 location), with mixing between two near-primitive magmas showing similar degrees of 2612 2613 1425 differentiation. Most importantly, well-constrained bubble-free MI that record evidence 2614 1426 of crystallization and mixing provide a chronological sequence of the magmatic 2615 2616 1427 processes occurring beneath Solchiaro. The recharging magma did not undergo extensive 2617 2618 1428 crystallization. In addition, well-constrained MI show no evidence for CO2 fluxing as 2619 1429 reported in other MI studies (e.g., Moretti et al., 2013). In fact, normal and bubble-free 2620 2621 1430 basaltic and trachybasaltic MI show that the H2O contents remained essentially constant 2622 1431 (within analytical error). In other words, well-constrained mafic MI show no evidence 2623 2624 1432 that H2O/CO2 decreases systematically with crystallization indicators (Fig. 17). Our 2625 2626 1433 results cannot disprove the CO2 fluxing interpretation, but neither do the MI data support 2627 1434 such an interpretation. 2628 2629 2630 47 2631 2632 2633 2634 2635 1435 2636 2637 1436 H2O and CO2 degassing during crystallization of trachytic magmas 2638 2639 1437 After considering bubble-free MI from basaltic and trachybasaltic magmas, we 2640 1438 investigated the bubble-free MI of more evolved magmas that likely originated from the 2641 2642 1439 differentiation of the same trachybasaltic melt. Ten bubble-free MI were analyzed in this 2643 1440 study, nine of them are representative of the NYT magma and the other is from the 2644 2645 1441 Solfatara eruption. Seven bubble-free MI are hosted in sanidine and the three are hosted 2646 2647 1442 in clinopyroxene. The low number of bubble-free MI analyzed is because bubble-free MI 2648 1443 large enough for SIMS analysis are rare in the studied phenocrysts. The bubble-free MI 2649 2650 1444 from the NYT and Solfatara samples were likely trapped during isobaric crystallization at 2651 1445 a relatively high pressure, ca. 200 MPa, followed by polybaric crystallization during 2652 2653 1446 ascent (Fig. 17c). This interpretation is consistent with the trend predicted by rhyolite- 2654 1447 MELTS, starting from Melt-1 composition as reported above. Isobaric crystallization of a 2655 2656 1448 trachybasaltic magma at ~200 MPa under volatile-saturated conditions assumed here is 2657 2658 1449 consistent with results of other studies (e.g., Arienzo et al., 2010). A melt zone at ~200 2659 1450 MPa is also suggested by geophysical data (e.g., Zollo et al. (2008), not only for CF and 2660 2661 1451 IP but for all of the Neapolitan volcanic area (see Moretti et al., 2013 and references 2662 1452 therein). While our interpretation is based on only a few well-constrained MI, we 2663

2664 1453 demonstrated that melts with high H2O contents can be generated during fractional 2665 2666 1454 crystallization, starting from trachybasaltic melts. The elevated H2O contents have 2667 1455 obvious and important implications for generating high-magnitude (explosive) eruptions. 2668 2669 1456 The composition of the trachytic bubble-free MI can be obtained after ~66% to 2670 1457 85% of isobaric crystallization under volatile-saturated conditions (Fig. 13b) (see also 2671 2672 1458 Appendix Table C). The mineral assemblage predicted by rhyolite-MELTS is consistent 2673 2674 1459 with the mineral assemblage of selected rock samples that are mainly composed of 2675 1460 clinopyroxene and feldspar ± biotite and Ti-magnetite (see Appendix Table C). We 2676 2677 1461 discriminated between trachytic bubble-free MI by host using ratios of elements 2678 1462 incompatible or slightly compatible with the host (Fig. 13c and d). For instance, bubble- 2679 2680 1463 free MI hosted in sanidine show good agreement with rhyolite-MELTS predictions and 2681 2682 1464 bulk rock trends for SiO2/CaO vs. MgO/TiO2 (Fig. 13c). Consistently, bubble-free MI 2683 1465 hosted in clinopyroxene show agreement with rhyolite-MELTS predictions and bulk rock 2684 2685 2686 48 2687 2688 2689 2690 2691 1466 trends for SiO / TiO vs. Al O /K O (Fig. 13d). It is important to note that trachytic 2692 2 2 2 3 2 2693 1467 bubble-free MI show good agreement even when ratios such as SiO2/CaO (CaO highly 2694 2695 1468 compatible with clinopyroxene) and Al2O3/K2O (K2O and Al2O3 compatible with 2696 1469 sanidine) are considered (Fig. 13b). These trends suggest that PEC or other modifications 2697 2698 1470 after trapping and before eruption were not extensive and did not significantly affect the 2699 1471 composition of the trapped melt. The inferred low degree of PEC is also consistent with 2700 2701 1472 agreement between compositions of MI and the bulk rocks, and with the occurrence of 2702 2703 1473 bubble-free MI (Steele-MacInnis et al., 2011). CO2 and H2O contents are available for 2704 1474 two bubble-free, trachytic MI from the NYT eruption. One bubble-free MI is hosted in 2705 2706 1475 clinopyroxene and the other in sanidine. The isobaric crystallization model is consistent 2707 1476 with the volatile contents of bubble-free MI hosted in clinopyroxene, but not with the 2708 2709 1477 bubble-free MI hosted in sanidine (Fig. 16c). In order to evolve a melt with the volatile 2710 1478 content recorded by the bubble-free MI hosted in sanidine, the system must crystallize at 2711 2712 1479 a lower pressure (shallower depth). Thus, our interpretation is that the last stage of 2713 2714 1480 crystallization for the NYT eruption occurs at shallower levels, compared to the earlier 2715 1481 stages of crystallization. Under volatile-saturated conditions, the H2O content of the melt 2716 2717 1482 can only increase if the magma evolves from trachybasalt to trachyte isobarically. With 2718 1483 this scenario, the CO content of the melt decreases and the melt becomes enriched in 2719 2

2720 1484 H2O and the fluid exsolving from the melt becomes progressively enriched in H2O (red 2721 2722 1485 dashed line in Fig. 8). The higher the pressure of isobaric crystallization, the higher the 2723 1486 H2O content of the evolved melts, as suggested by H2O-CO2-silicate melt solubility 2724 2725 1487 models. The higher the H2O content of more evolved melts the greater the likelihood of a 2726 1488 highly explosive eruption (e.g., NYT eruption), especially if crystallization occurs at 2727 2728 1489 relatively shallow depths where the difference between the partial molar volume of H2O 2729 2730 1490 dissolved in the melt and the molar volume of H2O-rich magmatic fluid is large (Bodnar 2731 1491 et al., 2007). In conclusion, some of the trachybasaltic magmas must undergo extensive 2732 2733 1492 isobaric crystallization (from 66% to 85%) to produce the relatively H2O-rich NYT-like 2734 1493 magmas. These magmas can migrate to a shallower level, triggering further exsolution of 2735 2736 1494 fluids and crystallization of sanidine. In agreement with our data and modeling, the NYT 2737 2738 1495 is the most explosive eruption (VEI 6) to occur in this area in the last 23.6 ka. This 2739 1496 interpretation is consistent with Mastrolorenzo and Pappalardo (2006), who showed that 2740 2741 2742 49 2743 2744 2745 2746 2747 1497 explosivity of eruptions at CF is mostly controlled by initial conditions in the magma 2748 2749 1498 chamber, including the volatile content, temperature and pressure. If the magnitude of the 2750 2751 1499 eruption is mainly driven by the pre-eruptive H2O content of melts (Edmonds and 2752 1500 Wallace, 2017; Webster et al., 2001), then extensive crystallization of a trachybasaltic 2753 2754 1501 magma occurring at relatively high pressure (relatively great depth) sets the stage for a 2755 1502 later possible highly explosive eruption, because it will generate a more evolved, H2O- 2756 2757 1503 rich melt. 2758 2759 2760 1504 Concluding remarks 2761 2762 1505 Major element compositions of MI trapped in the most abundant crystal phases (olivine, 2763 2764 1506 clinopyroxene, sanidine and plagioclase) precipitating from CF and IP magmas younger 2765 1507 than 23.6 ka, both from this study and from the literature, show compositional trends that 2766 2767 1508 to a large extent overlap with bulk rock compositions. Some MI show anomalous 2768 2769 1509 compositions, especially for Al2O3, P2O5, FeO, MgO, CaO and TiO2. These anomalous 2770 1510 MI can be explained through various processes, including grain scale dissolution– 2771 2772 1511 reaction–mixing (Danyushevsky et al., 2004), Fe loss and/or gain, and boundary layer 2773 1512 processes. Anomalous MI exhibit volatile concentrations that likely deviate significantly 2774 2775 1513 from actual melt compositions. For these reasons, we suggest that all MI analyzed for 2776 2777 1514 volatile concentrations must also be analyzed for major elements to test for the 2778 1515 representativeness of the trapped melt. In this study, all anomalous MI data were filtered 2779 2780 1516 from the dataset before interpreting volatile concentrations recorded by MI. 2781 1517 Another potential concern associated with determining the pre-eruptive volatile 2782 2783 1518 content of the original trapped melt based on MI analysis is the presence of a vapor 2784 2785 1519 bubble. Only a portion of the total volatile budget in the original trapped melt will be 2786 1520 accounted for if the volatile content of the bubble is not considered (Esposito et al., 2016 2787 2788 1521 and references therein), and interpretation of volatile contents based only on analysis of 2789 1522 the glass phase of bubble-bearing MI may lead to erroneous conclusions. Thus, volatile 2790 2791 1523 concentrations of “normal” and bubble-free MI are shown to be more reliable monitors of 2792 2793 1524 volatile evolution during differentiation from trachybasaltic to trachyte magmas younger 2794 1525 than 23.6 ka associated with the CF and IP eruptions. Major and trace element 2795 2796 1526 systematics, volatile contents, and petrological modeling based on “normal” and bubble- 2797 2798 50 2799 2800 2801 2802 2803 1527 free MI suggest that trachybasaltic magmas ascended through the crust from depths of at 2804 2805 1528 least ~7.5 km (~200 MPa). This depth is consistent with geophysical data reported by 2806 2807 1529 Zollo et al. (2008), who suggest a 1-2 km thick layer of partially molten rocks at a depth 2808 1530 of 8 km. At ≥ 7.5 km depth, trachybasalts can ascend through the crust to shallower 2809 2810 1531 levels while undergoing minimal crystallization under volatile saturation and pond to 2811 1532 feed eruptions associated with relatively less-differentiated magmas, as in the case of the 2812 2813 1533 Solchiaro eruption (23.6 ka ago). At 7.5 km depth, the same poorly differentiated 2814 2815 1534 magmas undergo extensive isobaric crystallization to form trachytic-phonolitic magmas 2816 1535 feeding eruptions such as the NYT (15 ka ago; Deino et al., 2004). During extensive 2817 2818 1536 isobaric crystallization, the magma could also assimilate 1%-2% crustal material 2819 1537 (D’Antonio et al., 2007). Isobaric crystallization at ~7 km depth of the parental melt of 2820 2821 1538 the Fondo Riccio eruption at CF was also suggested by Cannatelli (2012) based on MI 2822 1539 data and MELTS simulations. Extensive crystallization under volatile-saturated 2823

2824 1540 conditions increases the H2O content of the magma and, in turn, increases the likelihood 2825 2826 1541 of generating a high magnitude (explosive) eruption, as occurred for the NYT eruption 2827 1542 (40 km3 DRE). In another case, the trachybasaltic magmas may evolve to trachytic- 2828 2829 1543 phonolitic magmas at shallower levels and H2O enrichment of the magma during isobaric 2830 1544 crystallization would be limited. 2831 2832 1545 Bubble-free MI representative of magmas associated with the Solchiaro eruption 2833 2834 1546 suggest that mixing between trachybasaltic magmas and basaltic magmas occurred 2835 1547 beneath the IP. One group of bubble-free MI is associated with simple polybaric 2836 2837 1548 crystallization from a trachybasaltic primitive melt, while another group of MI define a 2838 1549 trend consistent with mixing between the trachybasaltic primitive magma and another 2839 2840 1550 basaltic primitive magma. The primitive basaltic magma shows a more moderate 2841 2842 1551 enrichment of elements compatible with a slab-derived fluid. The mixing occurred at 2843 1552 depths of at least ~6 km after simple crystallization of the trachybasaltic magma and 2844 2845 1553 likely did not affect the shallowest levels of the plumbing system. 2846 1554 Finally, even though interpretations based on volatile concentrations from MI 2847 2848 1555 may be sometimes problematic, they provide the only direct means to investigate 2849 2850 1556 magmatic processes such as degassing during magma ascent and crystallization. In this 2851 2852 2853 2854 51 2855 2856 2857 2858 2859 1557 study, we show that MI provide more reliable information concerning the pre-eruptive 2860 2861 1558 volatile contents of magmas if the following steps are taken: 2862 2863 1559 1- MI showing anomalous major element compositions relative to the bulk rock 2864 1560 trends are omitted from the interpretation. 2865 2866 1561 2- A detailed petrographic description of MI (e.g., size, shape, distance from the 2867 1562 host edge, presence of the bubble, presence of magnetite dust etc.) should be 2868 2869 1563 performed to rule out processes affecting the reliability of MI such as 2870 2871 1564 boundary layer effects, and diffusive loss or gain of volatiles after trapping. 2872 1565 3- Bubbles of bubble-bearing MI should be analyzed to check for the presence of 2873 2874 1566 volatiles. Recent studies show that up to 90% of the total original CO2 content 2875 1567 of an MI could be stored in the vapor bubble (Aster et al., 2016; Hartley et al., 2876 2877 1568 2014; Moore et al., 2015). For certain conditions, significant amounts of H2O 2878 1569 and S could also be lost into the bubble of MI (Esposito et al., 2016). 2879 2880 1570 4- Temporal relationships between MI studied and host phase should be assessed 2881 2882 1571 to test various interpretations defined by volatile trends. When available, 2883 1572 groups of MI trapped at the same time (Melt Inclusion Assemblages) should 2884 2885 1573 be examined to test the reliability of the data recorded by MI. 2886 1574 Cathodoluminescence mapping of quartz phenocrysts (Peppard et al., 2001) or 2887 2888 1575 phosphorous maps in olivine (Esposito et al., 2015) can be used to assess 2889 2890 1576 relative time of MI formation. Time constraints on MI formation can be used 2891 1577 to support or rule out interpretations on volatile trends recorded by MI. 2892 2893 1578 2894 2895 2896 2897 1579 Acknowledgments 2898 2899 1580 We would like to thank H.E. Belkin for his help and kind assistance with EMP 2900 1581 measurements, and N. Shimizu for assistance with SIMS analyses. In addition, we would 2901 2902 1582 like to thank G. Rolandi and A. Paone for their assistance during the sampling campaign. 2903 1583 Part of this project was funded by the PON-BIOENERGCHEM. CEM and RE 2904 2905 1584 acknowledge support from the Deep Carbon Observatory and National Science 2906 1585 Foundation Grant EAR 1732256. Support for RJB during the course of this investigation 2907 2908 1586 was provided by NSF grant EAR-1624589. Review comments by three anonymous 2909 2910 52 2911 2912 2913 2914 2915 1587 reviewers and A. Gómez-Tuena helped to identify some inconsistencies in earlier 2916 2917 1588 versions and led to a much improved manuscript. 2918 2919 1589 2920 1590 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 53 2967 2968 2969 2970 2971 2972 1591 References 2973 2974 1592 2975 1593 Acocella, V. and Funiciello, R., 1999. The interaction between regional and local tectonics during resurgent 2976 1594 doming: the case of the island of Ischia, Italy. Journal of Volcanology and Geothermal Research, 2977 1595 88(1): 109-123. 2978 1596 Acocella, V., Salvini, F., Funiciello, R. and Faccenna, C., 1999. The role of transfer structures on volcanic 2979 1597 activity at Campi Flegrei (Southern Italy). Journal of Volcanology and Geothermal Research, 2980 1598 91(2): 123-139. 1599 Albarède, F., 1996. Introduction to geochemical modeling. Cambridge University Press. 2981 1600 Albini, A., Cristofolini, R., Di Girolamo, P., Nardi, G., Rolandi, G. and Stanzione, D., 1977. Rare earth 2982 1601 element and thorium distribution in volcanic rocks of the potassic kindred from Procida Island and 2983 1602 the Phlegraen Fields, southern Italy. Atti Accademia Nazionale dei Lincei, Classe di Scienze 2984 1603 Fisiche Matematiche e Naturali, Rendiconti, 63(5): 416-429. 2985 1604 Alessio, M., Allegri, L., Azzi, C., Calderoni, G., Cortesi, C., Improta, S. and Petrone, V., 1989. 14C 2986 1605 tephrochronology with different fractions of Paleosol humic matter at Procida Island, Italy. 2987 1606 Radiocarbon, 31: 664-671. Alessio, M., Bella, F., Improta, S., Belluomini, G., Cortesi, C. and Turi, B., 1971. University of Rome 2988 1607 1608 carbon-14 dates IX. Radiocarbon, 13(2): 395-411. 2989 1609 Arienzo, I., Mazzeo, F., Moretti, R., Cavallo, A. and D'Antonio, M., 2016. Open-system magma evolution 2990 1610 and fluid transfer at Campi Flegrei caldera (Southern Italy) during the past 5ka as revealed by 2991 1611 geochemical and isotopic data: The example of the eruption. Chemical Geology, 427: 109- 2992 1612 124. 2993 1613 Arienzo, I., Moretti, R., Civetta, L., Orsi, G. and Papale, P., 2010. The feeding system of Agnano–Monte 2994 1614 Spina eruption (Campi Flegrei, Italy): Dragging the past into present activity and future scenarios. 2995 1615 Chemical Geology, 270(1–4): 135-147. 1616 Aster, E.M., Wallace, P.J., Moore, L.R., Watkins, J., Gazel, E. and Bodnar, R.J., 2016. Reconstructing CO 2996 1617 2 concentrations in basaltic melt inclusions using Raman analysis of vapor bubbles. Journal of 2997 1618 Volcanology and Geothermal Research, 323: 148-162. 2998 1619 Baker, D., 2008. The fidelity of melt inclusions as records of melt composition. Contributions to 2999 1620 Mineralogy and Petrology, 156(3): 377-395. 3000 1621 Barbieri, M., Di Girolamo, P., Locardi, E., Lombardi, G. and Stanzione, D., 1979. Petrology of the calc- 3001 1622 alkaline volcanics of the Parete 2 well (, Italy). Periodico di Mineralogia, 48: 53-74. 3002 1623 Beccaluva, L., Di Girolamo, P. and Serri, G., 1991. Petrogenesis and tectonic setting of the Roman 3003 1624 Volcanic Province, Italy. Lithos, 26: 191-221. 1625 Blundy, J. and Cashman, K., 2008. Petrologic reconstruction of magmatic system variables and processes. 3004 1626 Reviews in Mineralogy and Geochemistry, 69(1): 179-239. 3005 1627 Bodnar, R.J., Cannatelli, C., De Vivo, B., Lima, A., Belkin, H.E. and Milia, A., 2007. Quantitative model 3006 1628 for magma degassing and ground deformation (bradyseism) at Campi Flegrei, Italy; implications 3007 1629 for future eruptions. Geology, 35(9): 791-794. 3008 1630 Bodnar, R.J. and Student, J.J., 2006. Melt inclusions in plutonic rocks: Petrography and microthermometry. 3009 1631 In: J.D. Webster (Editor), Melt Inclusions in Plutonic Rocks. Mineralogical Association of 3010 1632 Canada, Montreal, Quebec, Canada, pp. 1-25. Brocchini, D., Principe, C., Castradori, D., Laurenzi, M.A. and Gorla, L., 2001. Quaternary evolution of the 3011 1633 1634 southern sector of the Campanian Plain and early Somma-Vesuvius activity: insights from the 3012 1635 1 well. Mineralogy and Petrology, 73(1-3): 67-91. 3013 1636 Bucholz, C.E., Gaetani, G.A., Behn, M.D. and Shimizu, N., 2013. Post-entrapment modification of 3014 1637 volatiles and oxygen fugacity in olivine-hosted melt inclusions. Earth and Planetary Science 3015 1638 Letters, 374(0): 145-155. 3016 1639 Cannatelli, C., 2012. Understanding magma evolution at Campi Flegrei (Campania, Italy) volcanic 3017 1640 complex using melt inclusions and phase equilibria. Mineralogy and Petrology, 104(1-2): 29-42. 3018 1641 Cannatelli, C., Doherty, A., Esposito, R., Lima, A. and De Vivo, B., 2016. Understanding a volcano 1642 through a droplet: A melt inclusion approach. Journal of Geochemical Exploration, 171: 4-19. 3019 3020 3021 3022 54 3023 3024 3025 3026 3027 1643 Cannatelli, C., Lima, A., Bodnar, R.J., De Vivo, B., Webster, J.D. and Fedele, L., 2007. Geochemistry of 3028 1644 melt inclusions from the Fondo Riccio and Minopoli 1 eruptions at Campi Flegrei (Italy). 3029 1645 Chemical Geology, 237: 418-432. 3030 1646 Capaldi, G., Gasparini, P., Moauro, A., Salvia, E. and Travaglione, O., 1972. Rare Earth abundances in the 3031 1647 alkaline volcanic rocks from Campania, South Italy. Earth and Planetary Science Letters, 17: 247- 3032 1648 257. 3033 1649 Cashman, K.V., Sparks, R.S.J. and Blundy, J.D., 2017. Vertically extensive and unstable magmatic 1650 systems: A unified view of igneous processes. Science, 355(6331): eaag3055. 3034 1651 Cecchetti, A., Fulignati, P., Marianelli, P., Proto, N. and Sbrana, A., 2001. The feeding system of Campi 3035 1652 Flegrei. Insights from melt and fluid inclusions on Ignimbrite Campana, Solchiaro and Minopoli 3036 1653 eruptions, GNV-INGV Meeting, Rome, Italy, pp. 190–191. 3037 1654 Cecchetti, A., Marianelli, P. and Sbrana, A., 2002-2003. L’eruzione di Astroni (caldera dei Campi Flegrei): 3038 1655 dati preliminari dallo studio di inclusioni silicatiche. Atti Soc. tosc. Sci. nat., Mem., 108: 59-63. 3039 1656 Chiodini, G., Vandemeulebrouck, J., Caliro, S., D'Auria, L., De Martino, P., Mangiacapra, A. and Petrillo, 3040 1657 Z., 2015. Evidence of thermal-driven processes triggering the 2005–2014 unrest at Campi Flegrei caldera. Earth and Planetary Science Letters, 414: 58-67. 3041 1658 1659 Cipriani, F., Marianelli, P. and Sbrana, A., 2008. Study of a pyroclastic sequence of Solfatara volcano 3042 1660 (Phlegrean Fields). Insights on volcanological and feeding system processes. Atti della Societa 3043 1661 Toscana di Scienze Naturali, Memorie Serie A, 113: 39-48. 3044 1662 Civetta, L., Carluccio, E., Innocenti, F., Sbrana, A. and Taddeucci, G., 1991. Magma chamber evolution 3045 1663 under the during the last 10 ka; trace element and isotope data. European 3046 1664 Journal of Mineralogy, 3(2): 415-428. 3047 1665 Civetta, L., Orsi, G., Pappalardo, L., Fisher, R.V., Heiken, G. and Ort, M., 1997. Geochemical zoning, 3048 1666 mingling, eruptive dynamics and depositional processes - The Campanian Ignimbrite, Campi 1667 Flegrei caldera, Italy. Journal of Volcanology and Geothermal Research, 75(3-4): 183-219. 3049 1668 D'Antonio, M., 2011. Lithology of the basement underlying the Campi Flegrei caldera: Volcanological and 3050 1669 petrological constraints. Journal of Volcanology and Geothermal Research, 200(1–2): 91-98. 3051 1670 D'Antonio, M., Civetta, L. and Di Girolamo, P., 1999a. Mantle source heterogeneity in the Campanian 3052 1671 Region (South Italy) as inferred from geochemical and isotopic features of mafic volcanic rocks 3053 1672 with shoshonitic affinity. . Mineralogy and Petrology 67: 163–192. 3054 1673 D'Antonio, M., Civetta, L., Orsi, G., Pappalardo, L., Piochi, M., Carandente, A., De Vita, S., Di Vito, M.A. 3055 1674 and Isaia, R., 1999b. The present state of the magmatic system of the Campi Flegrei caldera based 3056 1675 on a reconstruction of its behavior in the past 12 ka. Journal of Volcanology and Geothermal 1676 Research, 91(2-4): 247-268. 3057 1677 D'Antonio, M. and Di Girolamo, P., 1994. Petrological and geochemical study of mafic shoshonitic 3058 1678 volcanics from Procida-Vivara and Islands (Campanian Region, South Italy). Acta 3059 1679 Vulcanol, 5: 69-80. 3060 1680 D'Antonio, M., Tonarini, S., Arienzo, I., Civetta, L. and Di Renzo, V., 2007. Components and processes in 3061 1681 the magma genesis of the Phlegrean Volcanic District, southern Italy. Geological Society of 3062 1682 America Special Papers, 418: 203-220. 3063 1683 Da Toledo, P., Giacomo, 1539. Ragionamento del terremoto, del Nvovo monte, del aprimento di terra in Pozvolo nel anno 1538 : e dela significatione d'essi. Giouanni Sulztbah alemano, Napoli. 3064 1684 1685 Danyushevsky, L.V., Della-Pasqua, F.N. and Sokolov, S., 2000. Re-equilibration of melt inclusions trapped 3065 1686 by magnesian olivine phenocrysts from subduction-related magmas: petrological implications. 3066 1687 Contributions to Mineralogy and Petrology, 138(1): 68-83. 3067 1688 Danyushevsky, L.V., Leslie, R.A.J., Crawford, A.J. and Durance, P., 2004. Melt inclusions in primitive 3068 1689 olivine phenocrysts; the role of localized reaction processes in the origin of anomalous 3069 1690 compositions. Journal of Petrology, 45(12): 2531-2553. 3070 1691 Danyushevsky, L.V., McNeill, A.W. and Sobolev, A.V., 2002a. Experimental and petrological studies of 3071 1692 melt inclusions in phenocrysts from mantle-derived magmas: an overview of techniques, 1693 advantages and complications. Chemical Geology, 183(1–4): 5-24. 3072 1694 Danyushevsky, L.V. and Plechov, P., 2011. Petrolog3: Integrated software for modeling crystallization 3073 1695 processes. Geochemistry, Geophysics, Geosystems, 12(7): Q07021. 3074 1696 Danyushevsky, L.V., Sokolov, S. and Falloon, T.J., 2002b. Melt inclusions in olivine phenocrysts; using 3075 1697 diffusive re-equilibration to determine the cooling history of a crystal, with implications for the 3076 1698 origin of olivine-phyric volcanic rocks. Journal of Petrology, 43(9): 1651-1671. 3077 3078 55 3079 3080 3081 3082 3083 1699 De Astis, G., Pappalardo, L. and Piochi, M., 2004. Procida Volcanic History: new insights in the evolution 3084 1700 of the Phlegraean Volcanic District (Campania, Italy). Bulletin of Volcanology, 66: 622-641. 3085 1701 De Natale, G., Troise, C., Mark, D., Mormone, A., Piochi, M., Di Vito, M.A., Isaia, R., Carlino, S., Barra, 3086 1702 D. and Somma, R., 2016. The Campi Flegrei Deep Drilling Project (CFDDP): New insight on 3087 1703 caldera structure, evolution and hazard implications for the Naples area (Southern Italy). 3088 1704 Geochemistry, Geophysics, Geosystems, 17(12): 4836-4847. 3089 1705 de Vita, S., Orsi, G., Civetta, L., Carandente, A., D'Antonio, M., Deino, A., di Cesare, T., Di Vito, M.A., 1706 Fisher, R.V., Isaia, R., Marotta, E., Necco, A., Ort, M., Pappalardo, L., Piochi, M. and Southon, J., 3090 1707 1999. The Agnano–Monte Spina eruption (4100 years BP) in the restless Campi Flegrei caldera 3091 1708 (Italy). Journal of Volcanology and Geothermal Research, 91: 269-301. 3092 1709 De Vivo, B., Petrosino, P., Lima, A., Rolandi, G. and Belkin, H., 2010. Research progress in volcanology 3093 1710 in the Neapolitan area, southern Italy: a review and some alternative views. Mineralogy and 3094 1711 Petrology, 99(1): 1-28. 3095 1712 De Vivo, B., Rolandi, G., Gans, P.B., Calvert, A., Bohrson, W.A., Spera, F.J. and Belkin, H.E., 2001. New 3096 1713 constraints on the pyroclastic eruptive history of the Campanian volcanic plain (Italy). Mineralogy and Petrology, 73(1-3): 47-65. 3097 1714 1715 Deegan, F., Troll, V., Freda, C., Misiti, V., Chadwick, J.P., McLeod, C. and Davidson, J.P., 2010. Magma– 3098 1716 carbonate interaction processes and associated CO2 release at Merapi Volcano, Indonesia: insights 3099 1717 from experimental petrology. Journal of Petrology, 51(5): 1027-1051. 3100 1718 Deino, A.L., Orsi, G., De Vita, S. and Piochi, M., 2004. The age of the Neapolitan Yellow Tuff caldera- 3101 1719 forming eruption (Campi Flegrei Caldera, Italy) assessed by 40Ar/39Ar dating method. Journal of 3102 1720 Volcanology and Geothermal Research, 133: 157-170. 3103 1721 Di Girolamo, P., Ghiara, M.R., Lirer, Munno, R., Rolandi, G. and Stanzione, D., 1984. Vulcanologia e 3104 1722 petrologia dei Campi Flegreei. Volcanology and petrology of Phlegraean Fields. Bollettino della 1723 Società Geologica Italiana 103: 349-413. 3105 1724 Di Girolamo, P. and Stanzione, D., 1973. Lineamenti geologici e petrologici dell'isola di Procida. 3106 1725 Rendiconti della Societa Italiana di Mineralogia e Petrologia, 29: 81-125. 3107 1726 Di Renzo, V., Arienzo, I., Civetta, L., D'Antonio, M., Tonarini, S., Di Vito, M.A. and Orsi, G., 2011. The 3108 1727 magmatic feeding system of the Campi Flegrei caldera: Architecture and temporal evolution. 3109 1728 Chemical Geology, 281(3-4): 227-241. 3110 1729 Di Vito, M., Lirer, L., Mastrolorenzo, G., Rolandi, G. and Scandone, R., 1985. Volcanological map of 3111 1730 Campi Flegrei, Ministero Protezione Civile e Università di Napoli. 3112 1731 Di Vito, M.A., Arienzo, I., Braia, G., Civetta, L., D'Antonio, M., Di Renzo, V. and Orsi, G., 2011. The 1732 Averno 2 fissure eruption: a recent small-size explosive event at the Campi Flegrei Caldera (Italy). 3113 1733 Bulletin of Volcanology, 73: 295-320. 3114 1734 Di Vito, M.A., Isaia, R., Orsi, G., Southon, J., De Vita, S., D'Antonio, M., Pappalardo, L. and Piochi, M., 3115 1735 1999. Volcanism and deformation since 12,000 years at the Campi Flegrei caldera (Italy). Journal 3116 1736 of Volcanology and Geothermal Research, 91(2-4): 221-246. 3117 1737 Edmonds, M. and Wallace, P.J., 2017. Volatiles and exsolved vapor in volcanic systems. Elements, 13(1): 3118 1738 29-34. 3119 1739 Esposito, R., Badescu, K., Boyce, J.W. and Manning, C.E., 2015. Assessing the Origin of Olivines Based on Relative Age of Melt Inclusions. American Geophysical Union, Fall Meeting 2015, abstract 3120 1740 1741 volume: #V51D-3068. 3121 1742 Esposito, R., Bodnar, R.J., Danyushevsky, L., De Vivo, B., Fedele, L., Hunter, J., Lima, A. and Shimizu, 3122 1743 N., 2011. Volatile Evolution of Magma Associated with the Solchiaro Eruption in the Phlegrean 3123 1744 Volcanic District (Italy). Journal of Petrology, 52(12): 2431-2460. 3124 1745 Esposito, R., Hunter, J., Schiffbauer, J., Shimizu, N. and Bodnar, R.J., 2014. An assessment of the 3125 1746 reliability of melt inclusions as recorders of the pre-eruptive volatile content of magmas. American 3126 1747 Mineralogist, 99(5-6): 976-998. 3127 1748 Esposito, R., Klebesz, R., Bartoli, O., Klyukin, Y.I., Moncada, D., Doherty, A.L. and Bodnar, R.J., 2012. 1749 Application of the Linkam TS1400XY heating stage to melt inclusion studies. Central European 3128 1750 Journal of Geosciences, 4(2): 208-218. 3129 1751 Esposito, R., Lamadrid, H.M., Redi, D., Steele-MacInnis, M., Bodnar, R.J., Manning, C.E., De Vivo, B., 3130 1752 Cannatelli, C. and Lima, A., 2016. Detection of liquid H2O in vapor bubbles in reheated melt 3131 1753 inclusions: Implications for magmatic fluid composition and volatile budgets of magmas? 3132 1754 American Mineralogist, 101(7): 1691-1695. 3133 3134 56 3135 3136 3137 3138 3139 1755 Fanara, S., Botcharnikov, R.E., Palladino, D.M., Adams, F., Buddensieck, J., Mulch, A. and Behrens, H., 3140 1756 2015. Volatiles in magmas related to the Campanian Ignimbrite eruption: Experiments vs. natural 3141 1757 findings. American Mineralogist, 100(10): 2284-2297. 3142 1758 Faure, F. and Schiano, P., 2005. Experimental investigation of equilibration conditions during forsterite 3143 1759 growth and melt inclusion formation. Earth and Planetary Science Letters, 236(3-4): 882-898. 3144 1760 Fedele, F.G., Giaccio, B., Isaia, R., Orsi, G., Carroll, M. and Scaillet, B., 2007. The Campanian Ignimbrite 3145 1761 factor: towards a reappraisal of the Middle to Upper Palaeolithic ‘transition’. In: J. Grattan and R. 1762 Torrence (Editors), Living under the shadow: cultural impacts of volcanic eruptions. One World 3146 1763 Archeology. Left Coast Press, Walnut Creek (CA), pp. 19-41. 3147 1764 Fedele, L., Insinga, D., Calvert, A., Morra, V., Perrotta, A. and Scarpati, C., 2011. 40Ar/39Ar dating of tuff 3148 1765 vents in the Campi Flegrei caldera (southern Italy): toward a new chronostratigraphic 3149 1766 reconstruction of the Holocene volcanic activity. Bulletin of Volcanology, 73(9): 1323-1336. 3150 1767 Fourmentraux, C., Metrich, N., Bertagnini, A. and Rosi, M., 2012. Crystal fractionation, magma step 3151 1768 ascent, and syn-eruptive mingling; the Averno 2 eruption (Phlegraean Fields, Italy). Contributions 3152 1769 to Mineralogy and Petrology, 163(6): 1121-1137. Fowler, S.J., Spera, F.J., Bohrson, W.A., Belkin, H.E. and De Vivo, B., 2007. Phase equilibria constraints 3153 1770 1771 on the chemical and physical evolution of the Campanian Ignimbrite. Journal of Petrology, 48(3): 3154 1772 459-493. 3155 1773 Gebauer, S.K., Schmitt, A.K., Pappalardo, L., Stockli, D.F. and Lovera, O.M., 2014. Crystallization and 3156 1774 eruption ages of Breccia Museo (Campi Flegrei caldera, Italy) plutonic clasts and their relation to 3157 1775 the Campanian ignimbrite. Contributions to Mineralogy and Petrology, 167(1): 953. 3158 1776 Ghiara, M.R., Lirer, L. and Munno, R., 1979. Mineralogy and geochemistry of the "Low-Potassium Series" 3159 1777 of the Campania Volcanics (South Italy). Chemical Geology, 26: 29-49. 3160 1778 Ghiorso, M.S. and Gualda, G.A., 2015. An H2O–CO2 mixed fluid saturation model compatible with 1779 rhyolite-MELTS. Contributions to Mineralogy and Petrology, 169(6): 1-30. 3161 1780 Gilg, H., Lima, A., Somma, R., Belkin, H., De Vivo, B. and Ayuso, R., 2001. Isotope geochemistry and 3162 1781 fluid inclusion study of skarns from Vesuvius. Mineralogy and Petrology, 73(1-3): 145-176. 3163 1782 Gillot, P.Y., Chiesa, S., Pasquare, G. and Vezzoli, L., 1982. <33,000-yr K-Ar dating of the volcano-tectonic 3164 1783 horst of the Isle of Ischia, . Nature, 299(5880): 242-245. 3165 1784 Hartley, M.E., Maclennan, J., Edmonds, M. and Thordarson, T., 2014. Reconstructing the deep CO2 3166 1785 degassing behaviour of large basaltic fissure eruptions. Earth and Planetary Science Letters, 3167 1786 393(0): 120-131. 3168 1787 Hauri, E., Wang, J., Dixon, J.E., King, P.L., Mandeville, C. and Newman, S., 2002. SIMS analysis of 1788 volatiles in silicate glasses: 1. Calibration, matrix effects and comparisons with FTIR. Chemical 3169 1789 Geology, 183(1): 99-114. 3170 1790 Hayward, C., 2012. High spatial resolution electron probe microanalysis of tephras and melt inclusions 3171 1791 without beam-induced chemical modification. The Holocene, 22(1): 119-125. 3172 1792 Helo, C., Longpre, M.A., Shimizu, N., Clague, D.A. and Stix, J., 2011. Explosive eruptions at mid-ocean 3173 1793 ridges driven by CO2-rich magmas. Nature Geoscience, 4(4): 260-263. 3174 1794 Humphreys, M.C.S., Blundy, J.D. and Sparks, R.S.J., 2008a. Shallow-level decompression crystallisation 3175 1795 and deep magma supply at Shiveluch Volcano. Contributions to Mineralogy and Petrology, 155(1): 45-61. 3176 1796 1797 Humphreys, M.C.S., Menand, T., Blundy, J.D. and Klimm, K., 2008b. Magma ascent rates in explosive 3177 1798 eruptions: Constraints from H2O diffusion in melt inclusions. Earth and Planetary Science Letters, 3178 1799 270(1–2): 25-40. 3179 1800 Ippolito, F., D'Argenio, B., Pescatore, T. and Scandone, P., 1975. Structural-stratigraphic units and tectonic 3180 1801 framework of the Southern Apennines. In: C.H. Squyres (Editor), Geology of Italy. Earth Sci. Soc. 3181 1802 Libyan Arab Rep. , Tripoli Libya, pp. 1-11. 3182 1803 Isaia, R., Marianelli, P. and Sbrana, A., 2009. Caldera unrest prior to intense volcanism in Campi Flegrei 3183 1804 (Italy) at 4.0 ka BP: Implications for caldera dynamics and future eruptive scenarios. Geophysical 1805 Research Letters, 36(21). 3184 1806 Kent, A.J.R., 2008. Melt inclusions in basaltic and related volcanic rocks. In: K.D. Putirka and F.J. Tepley, 3185 1807 III (Editors), MInerals, Inclusions and Volcanic Processes. The Mineralogical Society of America, 3186 1808 pp. 273-331. 3187 1809 Klébesz, R., Esposito, R., De Vivo, B. and Bodnar, R.J., 2015. Constraints on the origin of sub-effusive 3188 1810 nodules from the Sarno (Pomici di Base) eruption of Mt. Somma-Vesuvius (Italy) based on 3189 3190 57 3191 3192 3193 3194 3195 1811 compositions of silicate-melt inclusions and clinopyroxene. American Mineralogist, 100(4): 760- 3196 1812 773. 3197 1813 Kress, V.C. and Ghiorso, M.S., 2004. Thermodynamic modeling of post-entrapment crystallization in 3198 1814 igneous phases. Journal of Volcanology and Geothermal Research, 137(4): 247-260. 3199 1815 Lima, A., Danyushevsky, L.V., De Vivo, B. and Fedele, L., 2003. A model for the evolution of the Mt. 3200 1816 Somma-Vesuvius magmatic system based on fluid and melt inclusion investigations. 3201 1817 Developments in Volcanology, 5: 227-249. 1818 Lima, A., De Vivo, B., Spera, F.J., Bodnar, R.J., Milia, A., Nunziata, C., Belkin, H.E. and Cannatelli, C., 3202 1819 2009. Thermodynamic model for uplift and deflation episodes (bradyseism) associated with 3203 1820 magmatic-hydrothermal activity at the Campi Flegrei (Italy). Earth-Science Reviews, 97(1-4): 44- 3204 1821 58. 3205 1822 Lima, A., Esposito, R. and De Vivo, B., 2017. Fluid and melt inclusions from subvolcanic to surface 3206 1823 environment in the Campi Flegrei (Napoli, Italy) active volcanic system. Journal of the Geological 3207 1824 Society of India, 90(5): 515-523. 3208 1825 Lirer, L., Petrosino, P., Alberico, I. and Armiero, V., 2011. Cartografia. In: L. Lirer (Editor), I Campi Flegrei, storia di un campo vulcanico. Quaderni dell'accademia pontiniana, Napoli, pp. 7-77. 3209 1826 1827 Lirer, L., Rolandi, G. and Rubin, M., 1991. 14C age of the "Museum Breccia" (Campi Flegrei) and its 3210 1828 relevance for the origin of the Campanian Ignimbrite. Journal of Volcanology and Geothermal 3211 1829 Research, 48(1-2): 223-227. 3212 1830 Maclennan, J., 2017. Bubble formation and decrepitation control the CO2 content of olivine‐hosted melt 3213 1831 inclusions. Geochemistry, Geophysics, Geosystems, 18(2): 597-616. 3214 1832 Mangiacapra, A., Moretti, R., Rutherford, M., Civetta, L., Orsi, G. and Papale, P., 2008. The deep 3215 1833 magmatic system of the Campi Flegrei caldera (Italy). Geophysical Research Letter, 35(21): 3216 1834 L21304. 1835 Marianelli, P., Sbrana, A. and Proto, M., 2006. Magma chamber of the Campi Flegrei at the 3217 1836 time of eruption of the Campanian Ignimbrite. Geology, 34(11): 937-940. 3218 1837 Marsh, B.D., 1996. Solidification fronts and magmatic evolution. Mineralogical Magazine, 60(1): 5-40. 3219 1838 Mastrolorenzo, G. and Pappalardo, L., 2006. Magma degassing and crystallization processes during 3220 1839 eruptions of high-risk Neapolitan-volcanoes: Evidence of common equilibrium rising processes in 3221 1840 alkaline magmas. Earth and Planetary Science Letters, 250: 164-181. 3222 1841 Mazzeo, F., D'Antonio, M., Arienzo, I., Aulinas, M., Di Renzo, V. and Gimeno, D., 2014. Subduction- 3223 1842 related enrichment of the Neapolitan Volcanoes (Southern Italy) mantle source: New constraints 3224 1843 on the characteristics of the slab-derived components. Chemical Geology, 386: 165-183. 1844 Melluso, L., De'Gennaro, R., Fedele, L., Franciosi, L. and Morra, V., 2012. Evidence of crystallization in 3225 1845 residual, Cl–F-rich, agpaitic, trachyphonolitic magmas and primitive Mg-rich basalt– 3226 1846 trachyphonolite interaction in the lava domes of the Phlegrean Fields (Italy). Geological 3227 1847 Magazine, 149(3): 532. 3228 1848 Metrich, N. and Wallace, P.J., 2008. Volatile abundances in basaltic magmas and their degassing paths 3229 1849 tracked by melt inclusions. Reviews in Mineralogy and Geochemistry, 69(1): 363-402. 3230 1850 Milia, A., 2010. The stratigraphic signature of volcanism off Campi Flegrei (Bay of Naples, Italy). In: G. 3231 1851 Groppelli and L. Viereck-Goette (Editors), Stratigraphy and geology of volcanic areas. The Geological Society of America, Buolder, Colorado, pp. 155-170. 3232 1852 1853 Mironov, N., Portnyagin, M., Botcharnikov, R., Gurenko, A., Hoernle, K. and Holtz, F., 2015. 3233 1854 Quantification of the CO2 budget and H2O–CO2 systematics in subduction-zone magmas through 3234 1855 the experimental hydration of melt inclusions in olivine at high H2O pressure. Earth and Planetary 3235 1856 Science Letters, 425: 1-11. 3236 1857 Mironov, N. and Portnyagin, M.V., 2011. H2O and CO2 in parental magmas of Kliuchevskoi volcano 3237 1858 inferred from study of melt and fluid inclusions in olivine. Russian Geology and Geophysics, 3238 1859 52(11): 1353-1367. 3239 1860 Moore, L.R., Gazel, E., Tuohy, R., Lloyd, A.S., Esposito, R., Steele-MacInnis, M., Hauri, E.H., Wallace, 1861 P.J., Plank, T. and Bodnar, R.J., 2015. Bubbles matter: An assessment of the contribution of vapor 3240 1862 bubbles to melt inclusion volatile budgets. American Mineralogist, 100(4): 806-823. 3241 1863 Morabito, S., Petrosino, P., Milia, A., Sprovieri, M. and Tamburrino, S., 2014. A multidisciplinary 3242 1864 approach for reconstructing the stratigraphic framework of the last 40ka in a bathyal area of the 3243 1865 eastern . Global and Planetary Change, 123: 121-138. 3244 3245 3246 58 3247 3248 3249 3250 3251 1866 Moretti, R., Arienzo, I., Orsi, G., Civetta, L. and D’Antonio, M., 2013. The Deep Plumbing System of 3252 1867 Ischia: a Physico-chemical Window on the Fluid-saturated and CO2-sustained Neapolitan 3253 1868 Volcanism (Southern Italy). Journal of Petrology, 54(5): 951-984. 3254 1869 Moretti, R., De Natale, G. and Troise, C., 2017. A geochemical and geophysical reappraisal to the 3255 1870 significance of the recent unrest at Campi Flegrei caldera (Southern Italy). Geochemistry, 3256 1871 Geophysics, Geosystems, 18(3): 1244-1269. 3257 1872 Morgan, G.B. and London, D., 2005. Effect of current density on the electron microprobe analysis of alkali 1873 aluminosilicate glasses. American Mineralogist, 90(7): 1131-1138. 3258 1874 Mormone, A., Piochi, M., Bellatreccia, F., De Astis, G., Moretti, R., Della Ventura, G., Cavallo, A. and 3259 1875 Mangiacapra, A., 2011. A CO2-rich magma source beneath the Phlegraean Volcanic District 3260 1876 (Southern Italy): Evidence from a melt inclusion study. Chemical Geology, 287(1-2): 66-80. 3261 1877 Moussallam, Y., Oppenheimer, C., Scaillet, B., Gaillard, F., Kyle, P., Peters, N., Hartley, M., Berlo, K. and 3262 1878 Donovan, A., 2014. Tracking the changing oxidation state of Erebus magmas, from mantle to 3263 1879 surface, driven by magma ascent and degassing. Earth and Planetary Science Letters, 393: 200- 3264 1880 209. Nakamura, M. and Shimakita, S., 1998. Dissolution origin and syn-entrapment compositional change of 3265 1881 1882 melt inclusion in plagioclase. Earth and Planetary Science Letters, 161(1): 119-133. 3266 1883 Norman, M.D., Garcia, M.O., Kamenetsky, V.S. and Nielsen, R.L., 2002. Olivine-hosted melt inclusions in 3267 1884 Hawaiian picrites: equilibration, melting, and plume source characteristics. Chemical Geology, 3268 1885 183(1–4): 143-168. 3269 1886 Orsi, G., Civetta, L., Dantonio, M., Digirolamo, P. and Piochi, M., 1995. Step-filling and development of a 3270 1887 3-layer magma chamber - the Neapolitan-Yellow-Tuff case-history. Journal of volcanology and 3271 1888 geothermal research, 67(4): 291-312. 3272 1889 Orsi, G., Civetta, L., Del Gaudio, C., De Vita, S., Di Vito, M.A., Isaia, R., Petrazzuoli, S.M., Ricciardi, 1890 G.P. and Ricco, C., 1999. Short-term ground deformations and seismicity in the resurgent Campi 3273 1891 Flegrei Caldera (Italy); an example of active block-resurgence in a densely populated area. Journal 3274 1892 of volcanology and geothermal research, 91(2-4): 415-451. 3275 1893 Orsi, G., D'Antonio, M., de Vita, S. and Gallo, G., 1992. The Neapolitan Yellow Tuff, a large-magnitude 3276 1894 trachytic phreatoplinian eruption; eruptive dynamics, magma withdrawal and caldera collapse. 3277 1895 Journal of volcanology and geothermal research, 53(1-4): 275-287. 3278 1896 Orsi, G., de Vita, S. and di Vito, M., 1996. The restless, resurgent Campi Flegrei nested caldera (Italy); 3279 1897 constraints on its evolution and configuration. Journal of volcanology and geothermal research, 3280 1898 74(3-4): 179-214. 1899 Papale, P., Moretti, R. and Barbato, D., 2006. The compositional dependence of the saturation surface of 3281 1900 H2O + CO2 fluids in silicate melts. Chemical Geology, 229(1-3): 78-95. 3282 1901 Pappalardo, L., Civetta, L., D'Antonio, M., Deino, A., Di Vito, M.A., Orsi, G., Carandente, A., De Vita, S., 3283 1902 Isaia, R. and Piochi, M., 1999. Chemical and Sr-isotopical evolution of the Phlegraean magmatic 3284 1903 system before the Campanian Ignimbrite and the Neapolitan Yellow Tuff eruptions. Journal of 3285 1904 volcanology and geothermal research, 91(2-4): 141-166. 3286 1905 Pappalardo, L., Piochi, M., D'Antonio, M., Civetta, L. and Petrini, R., 2002. Evidence for Multi-stage 3287 1906 Magmatic Evolution during the past 60kyr at Campi Flegrei (Italy) Dduced from Sr, Nd and Pb Isotope Data. Journal of Petrology, 43(8): 1415-1434. 3288 1907 1908 Peccerillo, A., 1999. Multiple mantle metasomatism in central-southern Italy; geochemical effects, timing 3289 1909 and geodynamic implications. Geology, 27(4): 315-318. 3290 1910 Peccerillo, A., 2017. The Campania Province, Cenozoic Volcanism in the Tyrrhenian Sea Region. 3291 1911 Advances in Volcanology. Springer International Publishing AG, Cham, Switzerland, pp. 159- 3292 1912 201. 3293 1913 Peppard, B.T., Steele, I.M., Davis, A.M., Wallace, P.J. and Anderson, A.T., 2001. Zoned quartz 3294 1914 phenocrysts from the rhyolitic Bishop Tuff. American Mineralogist, 86(9): 1034-1052. 3295 1915 Perrotta, A., Scarpati, C., Luongo, G. and Morra, V., 2006. The Campi Flegrei caldera boundary in the city 1916 of Naples. In: B. De Vivo (Editor), Developments in Volcanology. Elsevier, pp. 85-96. 3296 1917 Pescatore, T. and Rolandi, G., 1981. Osservazioni preliminari sulla stratigrafia dei depositi vulcanoclastici 3297 1918 nel settore SW dei Campi Flegrei. Preliminary observations of the stratigraphy of volcaniclastic 3298 1919 deposits of the south-west sector of Phlagrean Fields. Bollettino della Società Geologica Italiana, 3299 1920 100(2): 233-254. 3300 3301 3302 59 3303 3304 3305 3306 3307 1921 Piochi, M., Kilburn, C.R.J., Vito, M.A., Mormone, A., Tramelli, A., Troise, C. and Natale, G., 2013. The 3308 1922 volcanic and geothermally active Campi Flegrei caldera: an integrated multidisciplinary image of 3309 1923 its buried structure. International Journal of Earth Sciences: 1-21. 3310 1924 Piochi, M., Mastrolorenzo, G. and Pappalardo, L., 2005. Magma ascent and eruptive processes from 3311 1925 textural and compositional features of Monte Nuovo pyroclastic products, Campi Flegrei, Italy. 3312 1926 Bulletin of Volcanology, 67(7): 663-678. 3313 1927 Piochi, M., Polacci, M., De Astis, G., Zanetti, A., Mangiacapra, A., Vannucci, R. and Giordano, D., 2008. 1928 Texture and composition of pumices and scoriae from the Campi Flegrei caldera (Italy): 3314 1929 Implications on the dynamics of explosive eruptions. Geochem. Geophys. Geosyst., 9(3): Q03013. 3315 1930 Putirka, K.D., 2017. Down the Crater: Where Magmas are Stored and Why They Erupt. Elements, 13(1): 3316 1931 11-16. 3317 1932 Rhodes, J., 1981. Characteristics of primary basaltic magmas. Basaltic Volcanism Study Project: Basaltic 3318 1933 volcanism on the terrestrial planets. Pergamon Press, New York: 409-432. 3319 1934 Roach, A.L., 2005. The evolution of silicic magmatism in the post-caldera volcanism of the Phlegrean 3320 1935 Fields, Italy. Doctoral Thesis, Brown University, Providence. Roedder, E., 1979. Origin and significance of magmatic inclusions. Bulletin de Mineralogie, 102(5-6, 3321 1936 1937 Mineraux et Minerais): 487-510. 3322 1938 Rolandi, G., Bellucci, F., Heizler, M.T., Belkin, H.E. and De Vivo, B., 2003. Tectonic controls on the 3323 1939 genesis of ignimbrites from the Campanian Volcanic Zone, southern Italy. Mineralogy and 3324 1940 Petrology, 79(1): 3-31. 3325 1941 Rosi, M. and Sbrana, A., 1987. Campi flegrei. Quaderni de "La Ricerca Scientifica", 9: 175. 3326 1942 Rosi, M., Sbrana, A. and Vezzoli, L., 1988a. Correlazioni tefrostratigraphiche di alcuni livelli di Ischia, 3327 1943 Procida e Campi Flegrei. Memorie della Societa Geologica Italiana, 41: 1015-1027. 3328 1944 Rosi, M., Sbrana, A. and Vezzoli, L., 1988b. Stratigraphy of Procida and Vivara islands. Bollettino 1945 Nazionale Geofisica e Vulcanologia, 4: 500-525. 3329 1946 Rowe, M.C., Nielsen, R.L. and Kent, A.J.R., 2006. Anomalously high Fe contents in rehomogenized 3330 1947 olivine-hosted melt inclusions from oxidized magmas. American Mineralogist, 91(1): 82-91. 3331 1948 Scandone, R., 1997. Campi Flegrei, http://vulcan.fis.uniroma3.it/campi_flegrei/Campi_flegrei.html, pp. 3332 1949 Geologic map of Campi Flegrei modified after Di Vito et al. (1985). 3333 1950 Scandone, R., Bellucci, F., Lirer, L. and Rolandi, G., 1991. The structure of the Campanian Plain and the 3334 1951 activity of the Neapolitan volcanoes (Italy). Journal of Volcanology and Geothermal Research, 3335 1952 48(1-2): 1-31. 3336 1953 Scandone, R., Giacomelli, L. and Fattori Speranza, F., 2006. The volcanological history of the volcanoes of 1954 Naples: a review. In: B. De Vivo (Editor), Volcanism in the Campanian Plain Elsevier Sciences, 3337 1955 pp. 1-26. 3338 1956 Scarpati, C., Cole, P. and Perrotta, A., 1993. The Neapolitan Yellow Tuff - A large volume multiphase 3339 1957 eruption from Campi Flegrei, Southern Italy. Bulletin of Volcanology, 55: 343-356. 3340 1958 Self, S., 1976. The recent volcanology of Terceira, Azores. Journal of the Geological Society, 132(6): 645- 3341 1959 666. 3342 1960 Shishkina, T., Botcharnikov, R., Holtz, F., Almeev, R. and Portnyagin, M.V., 2010. Solubility of H2O- and 3343 1961 CO2-bearing fluids in tholeiitic basalts at pressures up to 500 MPa. Chemical Geology, 277(1): 115-125. 3344 1962 1963 Signorelli, S., Vaggelli, G., Francalanci, L. and Rosi, M., 1999. Origin of magmas feeding the Plinian 3345 1964 phase of the Campanian Ignimbrite eruption, Phlegrean Fields (Italy); constraints based on matrix- 3346 1965 glass and glass-inclusion compositions. Journal of Volcanology and Geothermal Research, 91(2- 3347 1966 4): 199-220. 3348 1967 Smith, V., Isaia, R. and Pearce, N., 2011. Tephrostratigraphy and glass compositions of post-15 kyr Campi 3349 1968 Flegrei eruptions: implications for eruption history and chronostratigraphic markers. Quaternary 3350 1969 Science Reviews, 30(25): 3638-3660. 3351 1970 Sobolev, A.V. and Danyushevsky, L.V., 1994. Petrology and Geochemistry of Boninites from the North 1971 Termination of the Tonga Trench: Constraints on the Generation Conditions of Primary High-Ca 3352 1972 Boninite Magmas. Journal of Petrology, 35(5): 1183-1211. 3353 1973 Sobolev, A.V., Dmitriev, L.V., Barsukov, V.L., Nevsorov, V.N. and Slutsky, A.B., 1980. The formation 3354 1974 conditions of the high magnesium olivines from the monomineralic fraction of Luna 24 regolith. 3355 1975 Proceedings of the Lunar and Planetary Science Conference, 11: 105-116. 3356 1976 Spera, F., 1980. Thermal evolution of plutons: a parameterized approach. Science, 207(4428): 299-301. 3357 3358 60 3359 3360 3361 3362 3363 1977 Steele-MacInnis, M., Esposito, R., Moore, L.R. and Hartley, M.E., 2017. Heterogeneously entrapped, 3364 1978 vapor-rich melt inclusions record pre-eruptive magmatic volatile contents. Contributions to 3365 1979 Mineralogy and Petrology, 172(4): 18. 3366 1980 Steele-MacInnis, M.J., Esposito, R. and Bodnar, R.J., 2011. Thermodynamic model for the effect of post- 3367 1981 entrapment crystallization on the H2O-CO2 systematics of volatile saturated silicate melt 3368 1982 inclusions. . Journal of Petrology, 52(12): 2461-2482. 3369 1983 Stock, M.J., Humphreys, M.C., Smith, V.C., Isaia, R. and Pyle, D.M., 2016. Late-stage volatile saturation 1984 as a potential trigger for explosive volcanic eruptions. Nature Geoscience, 9(3): 249-254. 3370 1985 Student, J.J. and Bodnar, R.J., 2004. Silicate melt inclusions in porphyry copper deposits; identification and 3371 1986 homogenization behavior. In: D.J. Kontak, A.J. Anderson and D.D. Marshall (Editors), The 3372 1987 Canadian Mineralogist, vol.42, Part 5. Mineralogical Association of Canada, Ottawa, pp. 1583- 3373 1988 1599. 3374 1989 Sun, S.S., 1980. Lead Isotopic Study of Young Volcanic-Rocks from Mid-Ocean Ridges, Ocean Islands 3375 1990 and Island Arcs. Philosophical Transactions of the Royal Society of London Series a- 3376 1991 Mathematical Physical and Engineering Sciences, 297(1431): 409-445. Sun, S.S. and McDonough, W.F., 1989. Chemical and isotopic systematics of oceanic basalts; implications 3377 1992 1993 for mantle composition and processes. Geological Society Special Publications, 42: 313-345. 3378 1994 Thomas, J.B. and Bodnar, R.J., 2002. A technique for mounting and polishing melt inclusions in small (>1 3379 1995 mm) crystals. American Mineralogist, 87(10): 1505-1508. 3380 1996 Thunell, R., Federman, A., Sparks, S. and Williams, D., 1979. Age, Origin, and Volcanological 3381 1997 Significance of the Y-5 Ash Layer in the Mediterranean. Quaternary Research, 12(2): 241-253. 3382 1998 Tonarini, S., D'Antonio, M., Di Vito, M.A., Orsi, G. and Carandente, A., 2009. Geochemical and B–Sr–Nd 3383 1999 isotopic evidence for mingling and mixing processes in the magmatic system that fed the Astroni 3384 2000 volcano (4.1–3.8 ka) within the Campi Flegrei caldera (southern Italy). Lithos, 107(3–4): 135-151. 2001 Tonarini, S., Leeman, W.P., Civetta, L., D'Antonio, M., Ferrara, G. and Necco, A., 2004. B/Nb and delta 3385 2002 11B systematics in the Phlegrean volcanic district, Italy. Journal of volcanology and geothermal 3386 2003 research, 133: 123-139. 3387 2004 Turi, B., Taylor, H.P., Jr. and Ferrara, G., 1991. Comparisons of 18O 16O and 87Sr/ 86Sr in volcanic rocks 3388 2005 from the Pontine Islands, M. Ernici, and Campania with other areas in Italy. Special Publication - 3389 2006 Geochemical Society, 3: 307-324. 3390 2007 Villemant, B., 1988. Trace element evolution in the Phlegrean Fields (Central Italy): fractional 3391 2008 crystallization and selective enrichment. Contributions to Mineralogy and Petrology, 98: 169-183. 3392 2009 Vitale, S. and Isaia, R., 2014. Fractures and faults in volcanic rocks (Campi Flegrei, southern Italy): Insight 2010 into volcano-tectonic processes. International Journal of Earth Sciences, 103(3): 801-819. 3393 2011 Webster, J.D., Raia, F., De Vivo, B. and Rolandi, G., 2001. The behavior of chlorine and sulfur during 3394 2012 differentiation of the Mt. Somma-Vesuvius magmatic system. Mineralogy and Petrology, 73(1-3): 3395 2013 177-200. 3396 2014 Webster, J.D., Raia, F., Tappen, C. and De Vivo, B., 2003. Pre-eruptive geochemistry of the ignimbrite- 3397 2015 forming magmas of the Campanian Volcanic Zone, Southern Italy, determined from silicate melt 3398 2016 inclusions. Mineralogy and Petrology, 79(1): 99-125. 3399 2017 Welsch, B., Faure, F., Famin, V., Baronnet, A. and Bachèlery, P., 2013. Dendritic Crystallization: A Single Process for all the Textures of Olivine in Basalts? Journal of Petrology, 54(3): 539-574. 3400 2018 2019 Wohletz, K., Orsi, G. and Devita, S., 1995. Eruptive Mechanisms of the Neapolitan-Yellow-Tuff 3401 2020 Interpreted from Stratigraphic, Chemical, and Granulometric Data. Journal of volcanology and 3402 2021 geothermal research, 67(4): 263-290. 3403 2022 Yaxley, G., Kamenetsky, V., Kamenetsky, M., Norman, M. and Francis, D., 2004. Origins of 3404 2023 compositional heterogeneity in olivine-hosted melt inclusions from the Baffin Island picrites. 3405 2024 Contributions to Mineralogy and Petrology, 148(4): 426-442. 3406 2025 Zollo, A., Maercklin, N., Vassallo, M., Dello Iacono, D., Virieux, J. and Gasparini, P., 2008. Seismic 3407 2026 reflections reveal a massive melt layer feeding Campi Flegrei Caldera. Geophysical Research 2027 Letters, 35: L12306. 3408 2028 3409 3410 2029 3411 3412 3413 3414 61 3415 3416 3417 3418 3419 2030 Tables 3420 3421 2031 3422 3423 2032 Table 1. Sample details and petrographic description of MI from this study. 3424 2033 3425 3426 2034 Table 2. Major element and volatile compositions of MI from this study. 3427 2035 3428

3429 2036 Table 3. Calculated Fomax predicted by rhyolite-MELTS compared with the Fomax 3430 3431 2037 reported by D’Antonio and Di Girolamo (1994). 3432 2038 3433 3434 2039 Table 4. MI and olivine host descriptions based on Melt-1 and Mixed groupping. 3435 2040 3436 3437 2041 Appendix Table A. GPS coordinates of samples collected for this study 3438 3439 2042 3440 2043 Appendix Table B. Standards used for EMPA. 3441 3442 2044 3443 2045 Appendix Table C. Results of Rhyolite-Melts calculations 3444 3445 2046 3446 3447 2047 Appendix Table D. Database of MI and bulk rock compositions of CF and IP eruptive 3448 2048 units. 3449 3450 2049 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 62 3471 3472 3473 3474 3475 2050 Figure captions 3476 3477 2051 3478 3479 2052 Figure 1. Simplified geological map of Campi Flegrei and Procida, modified after Di 3480 2053 Vito et al. (1985) by Scandone (1997). Locations of the caldera rims are based on Vitale 3481 3482 2054 and Isaia (2014), though we attribute the source of the CI to outside CF (De Vivo et al, 3483 2055 2001; Rolandi et al., 2003; De Natale et al, 2016). White circles with red outlines are 3484 3485 2056 locations of samples analyzed in this study (sample names indicated). NYT = Neapolitan 3486 3487 2057 Yellow Tuff; AMS = Agnano-Monte Spina; CI = Campanian Ignimbrite 3488 2058 3489 3490 2059 Figure 2. Major-element compositions of MI. The six panels represent plots of SiO2 3491 2060 concentration versus (a) CaO, (b) K O, (c) Na O, (d) TiO , (e) FeO and (f) P O . Large 3492 2 2 2 tot 2 5 3493 2061 symbols represent MI analyzed in this study (triangles = MI hosted in clinopyroxene; 3494 3495 2062 squares = MI hosted in sanidine; asterisks = MI hosted in plagioclase). Small symbols 3496 2063 represent data from the literature (triangles = MI hosted in clinopyroxene; squares = MI 3497 3498 2064 hosted in sanidine; diamonds = MI hosted in olivine; circles = MI hosted in biotite). Bulk 3499 2065 rock data from the literature are shown as stars (red stars = CF; black stars = IP) and 3500 3501 2066 xenoliths of CF are shown as x’s. The red-dashed line encircles the range in compositions 3502 3503 2067 of MI from CF, and the black-dashed line encircles the range in compositions of MI from 3504 2068 IP. Estimated uncertainties for the data from the present study are shown as error bars in 3505 3506 2069 the lower-right of each panel. See text for sources for the bulk rock and MI data from the 3507 2070 literature. 3508 3509 2071 3510 3511 2072 Figure 3. Volatile concentrations of MI versus SiO2 concentration for (a) Cl, (b) S, (c) 3512 2073 CO2, (d) H2O and (e) F. Symbols and fields are the same as in Figure 2, with the addition 3513 3514 2074 that bubble-free MI are differentiated using dark-blue symbols. Note that bulk rock data 3515 2075 show only low contents of Cl, S and F, whereas bulk rock data for H2O are absent. 3516 3517 2076 Estimated uncertainties for the data from the present study are shown as error bars in 3518 3519 2077 each panel. 3520 2078 3521 3522 2079 Figure 4. Volatile concentrations of MI. a) H2O versus CO2 showing full range of 3523 2080 reported CO2 concentrations; b) H2O versus CO2 showing MI with up to 1200 ppm CO2 3524 3525 3526 63 3527 3528 3529 3530 3531 2081 (red field encircles MI from the Neapolitan Yellow Tuff); c) F versus Cl. Note the 3532 3533 2082 positive correlation suggesting that F and Cl have similar behavior during the 3534 3535 2083 differentiation of magmas associated with CF and IP. Symbols and fields are the same as 3536 2084 in Figures 2 and 3. Estimated uncertainties for the data from the present study are shown 3537 3538 2085 as error bars in each panel. 3539 2086 3540 3541 2087 Figure 5. Major-element compositions of MI from this study and from the literature 3542 3543 2088 compared to bulk rock compositions. a) CaO vs. Al2O3 of MI discriminated by host phase 3544 2089 (symbols the same as in Figure 2); arrows point to plagioclase and clinopyroxene 3545 3546 2090 compositions of PVD. Note some less evolved MI showing high CaO and low Al2O3 3547 2091 contents. MI plotting off of the main bulk rock trend are hosted in clinopyroxene. b) CaO 3548 3549 2092 vs. FeOtot of MI discriminated based on host phase (symbols same as in Figure 2); note 3550 2093 that several MI hosted in clinopyroxene and olivine plot outside the trend defined by the 3551 3552 2094 bulk rock data. The red star represents the composition of clinopyroxene host associated 3553 3554 2095 with the overheated MI (RESE17-AD12-M2, from this study). The white filled circle 3555 2096 represents the hypothetical original trapped composition. Dashed lines represent trends 3556 3557 2097 associated with host dissolution, PEC, or Fe-loss by diffusion. c) CaO vs. Al2O3 of MI 3558 2098 discriminated as either naturally glassy (unheated) MI or MI reheated in the lab. It is 3559 3560 2099 important to note that the Ca-rich and Al-poor inclusions were all reheated in the lab. d) 3561 3562 2100 CaO vs. FeOtot of MI hosted in clinopyroxene, discriminated as either naturally glassy 3563 2101 (unheated) MI or MI reheated in the lab. It is important to note that heated MI are 3564 3565 2102 generally distributed to the right (higher CaO) side of the bulk rock trend. 3566 2103 3567 3568 2104 Figure 6. Major element compositions of mafic bulk rocks from Procida and 3569 3570 2105 crystallization trends predicted by Rhyolite-MELTS using starting compositions based on 3571 2106 bulk rocks or MI showing high Al2O3/K2O. a), b) and c): MgO vs. Al2O3/K2O; d), e) and 3572 3573 2107 f): SiO2/CaO vs. Al2O3/K2O. Symbols are for bulk rock compositions from Procida from 3574 2108 the literature: red-filled stars = Solchiaro eruption (De Astis et al., 2004; Di Girolamo et 3575 3576 2109 al., 1984); red-open stars = Vivara or Solchiaro eruptions (D'Antonio et al., 1999a); 3577 3578 2110 yellow crosses = Fiumicello eruption; purple cross = Pozzo Vecchio eruption (Di 3579 2111 Girolamo et al., 1984). Trends were calculated using rhyolite-MELTS (Gualda and 3580 3581 3582 64 3583 3584 3585 3586 3587 2112 Ghiorso, 2015) with three initial compositions: Apr22, Apr18 (bulk rocks; D’Antonio et 3588 3589 2113 al., 1999), or RESC5-O5-MA (MI; Esposito et al., 2011). Panels a) and d) show results 3590 3591 2114 obtained using the three initial compositions at constant pressure of 200 MPa and at NNO 3592 2115 oxidation state. Panels b) and e) show results obtained using Apr18 as the starting 3593 3594 2116 composition, the NNO redox buffer, and comparing isobaric versus polybaric 3595 2117 crystallization paths (see text for details). Panels c) and f) show results obtained using 3596 3597 2118 Apr22 as the initial composition, constant pressure of 200 MPa, and assuming oxidation 3598 3599 2119 states of NNO, QFM+2 and QFM-2. The trends were calculated assuming the melt is 3600 2120 saturated with respect to H2O-CO2. The calculated trends show a poor fit to the bulk rock 3601 3602 2121 data, suggesting that initial melts equivalent to the high Al2O3/K2O bulk rocks and MI 3603 2122 cannot generate the observed geochemical trends by crystal fractionation alone. 3604 3605 2123 3606 2124 Figure 7. Major element compositions of mafic bulk rocks from Procida and calculated 3607

3608 2125 crystallization trends using bulk rocks and MI showing low Al2O3/K2O. a), b) and c): 3609 3610 2126 MgO vs. Al2O3/K2O. d), e) and f): SiO2/CaO vs. Al2O3/K2O. Symbols are the same as in 3611 2127 Figure 6. Trends were calculated using rhyolite-MELTS (Gualda and Ghiorso, 2015) 3612 3613 2128 with two initial compositions: Ps3 (bulk rock; Di Girolamo et al., 1984) and RESC3-O9- 3614 2129 MA (MI; Esposito et al., 2011). Panels a) and d) show results obtained using the two 3615 3616 2130 initial compositions, a constant pressure of 200 MPa and at oxidation state equal to NNO. 3617 3618 2131 Panels b) and e) show results obtained using Ps3 as the starting composition, the NNO 3619 2132 buffer, and compare isobaric versus polybaric crystallization paths (see text for details). 3620 3621 2133 Panels c) and f) show results obtained using Ps3 as the initial composition, constant 3622 2134 pressure of 200 MPa, and oxidation states of NNO, QFM+2 and QFM-2. The trends were 3623 3624 2135 calculated assuming that the melt is saturated with respect to H2O-CO2. Note that trends 3625 3626 2136 show a good fit with the bulk rock data at low Al2O3/K2O ratios, suggesting that the low 3627 2137 Al2O3/K2O rocks and MI are compatible with trends predicted based on crystal 3628 3629 2138 fractionation. 3630 2139 3631 3632 2140 Figure 8. H2O-CO2 systematics of bubble-free MI, with calculated crystallization trends 3633 3634 2141 predicted by rhyolite-MELTS (Gualda and Ghiorso, 2015). Note that the bubble-free MI 3635 2142 shown exclude any MI deemed anomalous based on major-element compositions (see 3636 3637 3638 65 3639 3640 3641 3642 3643 2143 text for details). Symbols are the same as those in Figure 3, and uncertainties on each data 3644 3645 2144 point are the same as in Figure 4. Fractional crystallization trends are labelled as H2O- 3646 3647 2145 CO2 saturated isobaric versus polybaric, as well as H2O-CO2 undersaturated. Note that 3648 2146 the H2O-CO2 systematics of MI hosted in olivine are consistent with polybaric 3649 3650 2147 crystallization under H2O-CO2 saturated conditions. 3651 2148 3652 3653 2149 Figure 9. Compositions of clinopyroxene hosting MI from this study and clinopyroxene 3654 3655 2150 compositions predicted by rhyolite-MELTS (Gualda and Ghiorso, 2015). Panels a), b) 3656 2151 and c) compare measured clinopyroxene compositions with those predicted by the 3657 3658 2152 models using high-Al2O3/K2O bulk rocks and MI (Apr22, Apr18 and RESC5-05-MA; as 3659 2153 in Figure 6). Crosses are clinopyroxene analyses, and line styles are the same as in Figure 3660 3661 2154 6. Panels d), e) and f) compare measured clinopyroxene compositions with those 3662 2155 predicted by the models using low-Al O /K O bulk rocks and MI (Ps3 and RESC3-09- 3663 2 3 2 3664 2156 MA; as in Figure 7). Line styles are the same as in Figure 7. Note that trends calculated 3665 3666 2157 using Ps3 as starting composition and polybaric fractional crystallization (panel e) show 3667 2158 the best fit. 3668 3669 2159 3670 2160 Figure 10. Compositions of feldspars hosting MI analyzed in this study (cross symbols) 3671 3672 2161 and from the literature (encircled fields), compared to feldspar compositions predicted by 3673 3674 2162 rhyolite-MELTS (trend lines; Gualda and Ghiorso, 2015). Panels a), b) and c) compare 3675 2163 measured feldspar compositions and those predicted by the models using high-Al2O3/K2O 3676 3677 2164 bulk rocks and MI (Apr22, Apr18 and RESC5-05-MA; as in Figure 6). Trend line styles 3678 2165 are the same as in Figure 6. Panels d), e) and f) compare measured feldspar compositions 3679 3680 2166 and those predicted by the models using low-Al2O3/K2O bulk rocks and MI (Ps3 and 3681 3682 2167 RESC3-09-MA; as in Figure 7). Line styles are the same as in Figure 7. Note that the best 3683 2168 fit is calculated using Ps3 as starting composition (see panels d), e) and f)). 3684 3685 2169 3686 2170 Figure 11. TAS diagram comparing MI from CF and IP and liquid compositions 3687 3688 2171 predicted by Rhyolite-MELTS. Panels a), b) and c) compare MI compositions and liquid 3689 3690 2172 compositions predicted by the models using high-Al2O3/K2O bulk rocks and MI (Apr22, 3691 2173 Apr18 and RESC5-05-MA; as in Figure 6). Trend line styles are the same as in Figure 6. 3692 3693 3694 66 3695 3696 3697 3698 3699 2174 Panels d), e) and f) compare MI compositions and liquid composition predicted by the 3700 3701 2175 models using low-Al2O3/K2O bulk rocks and MI (Ps3 and RESC3-09-MA; as in Figure 3702 3703 2176 7). Line styles are the same as in Figure 7. In all panels, the appearance and 3704 2177 disappearance of olivine, clinopyroxene, and sanidine predicted by rhyolite-MELTS are 3705 3706 2178 highlighted. Note the good fit between the phase field defined by the MI and their hosts 3707 2179 and rhyolite-MELTS predictions, relative to the appearance and disappearance of olivine, 3708 3709 2180 clinopyroxene, and sanidine, especially for panels d), e) and f). Data references for MI 3710 3711 2181 are as in the main text. 3712 2182 3713 3714 2183 Figure 12. Fo mol% of host olivine versus the respective Al2O3/K2O ratios of olivine- 3715 2184 hosted MI, compared with rhyolite-MELTS calculations assuming 200 MPa isobaric 3716 3717 2185 conditions. Symbols are as in Figure 3, while trend line styles are as in Figures 6 and 7. 3718 2186 Note that for all calculations Al O /K O of liquids does not change significantly during 3719 2 3 2 3720 2187 crystallization, whereas the bubble-free MI show a large variation 3721 3722 2188 3723 2189 Figure 13. Ratios of oxide concentrations (wt%), comparing bubble-free MI, bulk rock 3724 3725 2190 compositions and rhyolite-MELTS modeling results. The MI data exclude MI deemed 3726 2191 anomalous based on major-element concentrations (see text for details). Symbols are the 3727 3728 2192 same as in Figures 2 and 3, and trend line styles are the same as in Figures 6 and 7. a) 3729 3730 2193 SiO2/CaO vs. Al2O3/K2O of basaltic and trachybasaltic bubble-free MI hosted in olivine. 3731 2194 b) SiO2/CaO vs. Al2O3/K2O including trachytic bubble-free MI (symbols indicate host 3732 3733 2195 mineral). c) SiO2/CaO vs. MgO/TiO2 of bubble-free MI hosted in sanidine. d) SiO2/TiO2 3734 2196 vs. Al O /K O of bubble-free MI hosted in clinopyroxene. 3735 2 3 2 3736 2197 3737 3738 2198 Figure 14. Trace element compositions of bubble-free MI hosted in olivine. Symbols are 3739 2199 the same as in Figures 2 and 3. a) SiO2/CaO vs. Al2O3/ Rb, b) Ba/Sr vs. Yb/Rb, and c) 3740 3741 2200 Ba/Sr vs. Eu/Rb. Note that for trace elements, fewer bulk rock data are available 3742 2201 compared to major element data. The green arrow on each panel indicates the expected 3743 3744 2202 trend for polybaric crystallization. The red line on each panel represents a calculated 3745 3746 2203 mixing trend based on the same MI end members for all three panels. Mixing calculations 3747 2204 were based on a simple binary mixing (Albarède, 1996) 3748 3749 3750 67 3751 3752 3753 3754 3755 2205 3756 3757 2206 Figure 15. Spider diagrams of the MI and bulk rocks. Symbols in a) and b) are the same 3758 3759 2207 as in Figures 2 and 3. a) Incompatible trace element spider diagram with normalization to 3760 2208 primitive mantle as in (Sun, 1980), and b) Rare Earth elements spider diagram with 3761 3762 2209 normalization to chondrite as in Sun and McDonough (1989). c) Incompatible trace 3763 2210 element spider diagram of two MI: RESC3-O9-MA and RESC5-05-MA. These two MI 3764 3765 2211 are hosted in olivine showing similar Fo-rich compositions, but these MI show very 3766 3767 2212 different Al2O3/K2O ratios. Note that RESC3-O9-MA was selected as the low-Al2O3/K2O 3768 2213 starting composition for rhyolite melts simulations, whereas RESC5-05-MA was selected 3769 3770 2214 as the high-Al2O3/K2O starting composition for these simulations, shown in Figures 6 to 3771 2215 13. 3772 3773 2216 3774 2217 Figure 16. Ratios of volatile, trace and major element compositions for bubble-free MI. 3775

3776 2218 Symbols the same as in Figure 3. a) Cl/Yb vs. Al2O3/K2O; b) TiO2/S vs. Al2O3/Rb. Note 3777 3778 2219 that including Cl and S in these element ratio diagrams still differentiates the trends 3779 2220 formed by the two different groups of MI as seen in Figure 14. The arrows indicate trends 3780 3781 2221 predicted for polybaric fractional crystallization versus mixing. 3782 2222 3783 3784 2223 Figure 17. Volatile compositions of bubble-free MI compared to a) host compositions, 3785 3786 2224 and b) and c) major element compositions of the MI. Symbols the same as in Figure 3, 3787 2225 whereas trend line styles are the same as in Figures 6 and 7. a) mol% Fo vs. H2O/CO2; b) 3788 3789 2226 Al2O3/K2O vs. H2O/CO2 for mafic bubble-free MI; c) Al2O3/K2O vs. H2O/CO2 trachytic 3790 2227 and mafic bubble-free MI. 3791 3792 2228 3793 3794 2229 Figure 18. Simplified temporal evolution model for the Solchiaro eruption based on 3795 2230 bubble-free MI hosted in olivine. The results are based on the column-mush model of 3796 3797 2231 Marsh (1996), in which the plumbing system is represented by numerous sill-like pools 3798 2232 of magma variably connected by dikes. The different thicknesses of these sills and dikes 3799 3800 2233 can result in different P-T histories for the same type of melt. This is consistent with MI 3801 3802 2234 data that show no correlation between Fo mol% of the host and the extrapolated depths of 3803 2235 formation. Calculated pressures of formation of MI and interpreted depths were reported 3804 3805 3806 68 3807 3808 3809 3810 3811 2236 by Esposito et al. (2011) based on the solubility model of Papale et al. (2006). (t ) the 3812 1 3813 2237 magma evolves mainly by fractional crystallization of magmas in separate pools. (t2) 3814 3815 2238 recharge of new magma that mixes with the pre-existing magmas leading to the Solchiaro 3816 2239 eruption. 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 69 3863 3864 REPR11 NYT Monte Nuovo crater Vivara (>40 ka)

Solchiaro (23.6 ka) RESE17 Fossa Lupara Campanian Ignimbrite (39 ka)

RESO8 products older than NYT Solfatara (>15 ka) North Neapolitan Yellow Tuffs (~15 ka)

Lake Volcanoes of yellow tuffs (~12-8 ka) REBA1 Solfatara Bacoli REMS12 crater AMS Pyroclastic fall deposits (< ~8 ka)

Gulf of Pozzuoli Base surge deposits Monte di Procida (< ~8 ka)

Pyroclastic flow deposits (Monte Nuovo and Monte Spina) Island of Procida Lava domes

Volcano-sedimentary deposits RESC2 NYT caldera Faults Solchiaro 3 Km (Vitale and Isaia, 2014) CI caldera Starza marine Solchiaro Bay (Vitale and Isaia, 2014) terrace Fig. 1  This study:  MI hosted in cpx a MI hosted in san d MI hosted in plg  MI of CF MI of IP Literature:  MI hosted in cpx MI hosted in san

 ZW

MI hosted in oli  MI hosted in bio Bulk rocks of CF 7L2 &D2ZW Bulk rocks of IP  error bars  Xenoliths of CF error bars

             

6L2ZW 6L2ZW   b e 

 

 2ZW  .

 )H2WRWZW  error bars error bars 

             

6L2ZW 6L2ZW   c f

 

 ZW 2ZW   2  1D 3   error bars error bars

             

6L2ZW 6L2ZW Fig.2 V2 Bubble-free MI hosted in oli  Bubble-free MI hosted in cpx  Bubble-free MI hosted in san a d       2ZW  &OSSP  +  error bars  error bars  

              6L2 ZW  6L2ZW   b e     6SSP )SSP   error bars   error bars              

6L2ZW 6L2ZW  c

 SSP 

&2  error bars

      

6L2ZW Fig. 3 Rev2  a

 SSP 

&2 error bars 

        

+2ZW  b 

 NYT MI SSP

  &2  error bars 

        

+2ZW  c     )SSP   error bars         &OSSP

Fig. 2 Rev2 30 30 a unheated MI c plagioclase heated MI 25 25 wt% wt% 20 20 3 3 O O 2 2 Al Al 15 15

error bars cpx cpx 10 10 0 5 10 15 0 5 10 15 CaO wt% CaO wt% 15 15 b unheated MI hosted in cpx d overheated heated MI hosted in cpx 10 PEC MI 10 PEC overheated evolving cpx MI wt% wt% evolving cpx tot tot composition composition FeO FeO host host 5 dissolution 5 dissolution

di usive cpx host di usive Fe-loss Fe-loss 0 0 0 5 10 15 20 25 0 5 10 15 20 25 CaO wt% CaO wt% Fig. 5 Rev2 200 MPa and NNO FC trends initial composition test BR Solchiaro initial composition 200 MPa and NNO BR Solchiaro-Vivara Apr22  BR Fiumicello  initial composition a d Apr18  BR Pozzo Vecchio initial composition Apr22 Apr18 RESC5-O5-MA   RESC5-O5-MA

&D2  

0J2ZW  6L2  

    $O 2 . 2    $O2.2 polybaric vs. isobaric test Apr18 and NNO Apr18 and NNO FC trends   isobaric b e  polybaric  

&D2  polybaric  6L2 0J2ZW    isobaric     $O 2 . 2    $O2.2 Oxidation state test Apr22 and 200 MPa FC trends Apr22 as initial and 200 MPa NNO   QFM +2 c f  QFM -2   QFM +2

&D2  

0J2ZW QFM -2  6L2   NNO    

$O2.2 $O 2 . 2 Fig. 6 Rev2    Melt-1 initial composition test 200 MPa and NNO

 200 MPa and NNO FC trends  a initial composition RESC3-O9-MA  RESC3-O9-MA RESC3-O9-MA Ps3  initial composition  Ps3

&D2   Ps3

0J2ZW  6L2   d     $O 2 . 2    $O2.2 polybaric vs. isobaric test Ps3 as initial composition and NNO  Ps3 and NNO FC trends b  isobaric  polybaric  0.5 °C/MPa  polybaric

polybaric &D2   0.1 °C/MPa 6L2 0J2ZW    isobaric e     $O 2 . 2    $O2.2 Oxidation state test Ps3 as initial composition and 200 MPa  Ps3 and 200 MPa FC trends  c NNO  QFM +2 QFM +2   QFM -2

&D2  

0J2ZW  6L2   QFM -2 f    

$O2.2 $O 2 . 2 Fig. 7 Rev2    

2

O-CO isobaric xtl H 2

unsaturated

SSP  polybaric xtl 

H2O-CO2

&2 saturated isobaric xtl H 2 O-CO saturated2

 

+2ZW Fig. 8 Rev2 Mixed melt initial composition test Melt-1 initial composition test

200 MPa and NNO 200 MPa and NNO 200 MPa and NNO FC trends initial composition :R :R (a) 200 MPa and NNO FC trends (d) RESC3-O9-MA

initial composition initial composition Apr22 Ps3 initial composition 50 Apr18 50 initial composition RESC5-O5-MA

40 40 60 60

polybaric vs. isobaric test polybaric vs. isobaric test (Q )V (Q )V Apr18 and NNO Ps3 and NNO Ps3 and NNO FC trends

:R :R isobaric polybaric (b) Apr18 and NNO FC trends (e) 0.5 °C/MPa

isobaric polybaric 50 50 0.1 °C/MPa polybaric

40 40 60 60

(Q )V (Q )V Oxidation state test Oxidation state test Apr22 and 200 MPa Ps3 and 200 MPa

:R :R Ps3 and 200 MPa FC trends (c) Apr22 and 200 MPa FC trends (f) NNO NNO QFM +2 50 QFM +2 50 QFM -2 QFM -2

40 40 60 60

)V (Q )V (Q Fig. 9 Rev2 Mixed melt Melt-1 initial composition test initial composition test 200 MPa and NNO 200 MPa and NNO (a) (d) $Q $Q Feldspar this study

200 MPa and NNO FC trends 200 MPa and NNO FC trends initial composition initial composition xtl RESC3-O9-MA Apr22 xtl initial composition initial composition Apr18 Ps3 initial composition RESC5-O5-MA coexisting felds $E 2U $E 2U xtl xtl

polybaric vs. isobaric test polybaric vs. isobaric test Apr18 and NNO Ps3 and NNO (b) (e) $Q $Q Apr18 and NNO FC trends Ps3 and NNO FC trends

isobaric isobaric polybaric polybaric xtl xtl 0.5 °C/MPa polybaric 0.1 °C/MPa

coexisting felds $E 2U $E 2U xtl xtl

Oxidation state test Oxidation state test Apr22 and 200 MPa Ps3 and 200 MPa (c) (f) $Q Apr22 and 200 MPa FC trends $Q

NNO Ps3 and 200 MPa FC trends QFM +2 NNO QFM -2 xtl xtl QFM +2 QFM -2

coexisting felds $E 2U $E 2U xtl Fig. 10 Rev2 xtl Mixed melt Melt-1 initial composition test initial composition test 200 MPa and NNO 200 MPa NNO

  a 3KRQROLWH cpx-oli out 3KRQROLWH san in d cpx-oli out san in  7HSKUL  7HSKUL SKRQROLWH SKRQROLWH   7UDFK\WH cpx in 7UDFK\WH oli out san in  cpx out  cpx in oli in

oli out 2ZW 2ZW     2. 2.   /DWLWH /DWLWH 

 1D 1D 6KRVKRQLWH 200 MPa and NNO FC trends 6KRVKRQLWH 200 MPa and NNO FC trends 7UDFK\EDVDOW initial composition 7UDFK\EDVDOW   Apr22 oli in cpx in initial composition initial composition Ps3  oli in  %DVDOW %DVDOW cpx in Apr18 %DVDOW oli in %DVDOW initial composition initial composition RESC3-O9-MA   RESC5-O5-MA                   6L2ZW 6L2ZW  polybaric vs. isobaric test polybaric vs. isobaric test Apr18 and NNO Ps3 and NNO  san in  b 3KRQROLWH cpx-oli out 3KRQROLWH san in san in e oli out  7HSKUL  7HSKUL SKRQROLWH SKRQROLWH  cpx in  7UDFK\WH cpx in 7UDFK\WH

  oli in 2ZW 2ZW     2. 2.   /DWLWH /DWLWH   1D 1D 6KRVKRQLWH 6KRVKRQLWH 7UDFK\EDVDOW 7UDFK\EDVDOW Ps3 and NNO FC trends  Apr18 and NNO FC trends  oli in cpx in isobaric  oli in %DVDOW cpx in isobaric  %DVDOW polybaric %DVDOW oli in 0.5 °C/MPa  polybaric  polybaric                 0.1 °C/MPa 

6L2ZW 6L2ZW Oxidation state test Oxidation state test Apr22 and 200 MPa Ps3 and 200 MPa QFM-2 QFM-2 NNO san in NNO san in   c cpx-oli out 3KRQROLWH san in 3KRQROLWH QFM+2 f san in  7HSKUL QFM+2  7HSKUL SKRQROLWH SKRQROLWH  cpx out  7UDFK\WH cpx in 7UDFK\WH cpx in  san in  cpx in cpx-oli out oli in 2ZW 2ZW     2.

2. oli in   /DWLWH /DWLWH   1D 1D 6KRVKRQLWH 6KRVKRQLWH 7UDFK\EDVDOW 7UDFK\EDVDOW  Apr22 and 200 MPa FC trends  Ps3 and 200 MPa FC trends cpx in oli in cpx in  oli in %DVDOW NNO  %DVDOW oli in %DVDOW NNO          QFM +2       QFM  +2  6L2ZW 6L2ZW  QFM -2  QFM -2 Fig. 11 Rev2  a Apr22 Apr18 corrected MI

2 (RESC5-O5-MA) 

.  bubble-free MI hosted  in olivine from th eliterature 2 

$O Ps3 corrected MI  (RESC3-O9-MA)

   +RVW)R PRO Fig. 12 Rev2  a  bubble-free MI hosted in olivine from  the literature BR of IP &D2

 

6L2  

 

$O2.2 50 b 40 Bubble-free MI hosted in oli Bubble-free MI hosted in cpx Bubble-free MI hosted in san 30 Bulk rocks of CF Bulk rocks of IP

&D2 Xenoliths of CF  20 6L2 10

0 0 5 10 15

$O2.2 50 c 40

30 &D2  20 6L2 10

0 

0J27L2  d

 

7L2   6L2 

 0 5 10 15

Fig. 13 Rev2 $O2.2 9 a bubble-free MI hosted 8 in olivine from the literature 7 BR of IP

/CaO 6

2 polybaric xlt polybaric Mixing

SiO 5 8060 20 4 40 3 0 1000 2000 3000 4000 5000 Al O /Rb 2.2 2 3 b 2.0 1.8

1.6 polybaric xlt polybaric

Ba/Sr 1.4 1.2 Mixing 1.0 80 60 40 20 0.8 0.00 0.02 0.04 0.06 0.08 Yb/Rb 2.2 c 2.0 1.8 1.6 1.4

Ba/Sr polybaric xlt polybaric 1.2 40 Mixing 80 60 1.0 20 0.8 0.00 0.02 0.04 0.06 0.08 0.10 Eu/Rb

Fig. 14 Rev2 a





 5RFN3ULPLWLYH0DQWOH

 &V 5E %D 7K 8 1E 7D . /D&H3E3U 6U 3 1G =U 6P (X7L '\ <

b



 5RFN&KRQGULWHV

 /D &H 3U 1G 3P 6P (X *G 7E '\ +R (U 7P

hosted in Fo~88 

 RESC5-O5-MA High Al/K MI 5RFN3ULPLWLYH0DQWOH hosted in Fo~87  &V 5E %D 7K 8 1E 7D . /D&H3E3U 6U 3 1G =U 6P (X7L '\ <

2000

Cl/Yb Mixing

1000 fractionation polybaric

0 0 5 10 15 20

Al2O3/K2O 18 16 b 14

/S 12 2

TiO 10 Mixing 8

6 polybaric fractionation 4 0 1000 2000 3000 4000 Fig.16 Rev2 Al2O3/Rb 140 a 120 100

2

0.1 °C/MPa 0.1 80 °C/MPa 0.5 O/CO

2 60 H 40 20 0 80 85 90 Fo mol% 140 b

120

t

l x

100

c i

2

r a

80 b

y

l o

O/CO p

2 60 xlt isobaric H 40 20 0 0 5 10 15

$O2.2

240 t c

l

x

c

i

200 r

a

b

y 2

l polybaric xlt

160 o

p 120 O/CO 2 H 80

40 xlt isobaric 0 0 5 10 15 Fig.17 Rev2 $O2.2

Fig. 18 Rev2 18 Fig.

hosted in olivine in hosted hosted in olivine in hosted

mush zone mush

Mixed, bubble-free MI bubble-free Mixed, Melt-1, bubble-free MI bubble-free Melt-1,

200 200 7.4 7.4 7.4 7.4 Pressure (MPa)

Depth (km)

Depth (km)

3.7 100 100 3.7

1

2 (b) t (a) t 0

0 0 0 Figure captions for supplementary figures

Supplementary Figure 1. Bulk rock compositions from CF and IP. (a) TAS diagram (LeBas et al., 1986) showing the CF and IP bulk rock (Na2O+K2O) versus SIO2 relationships. (b) Concentration of MgO (wt%) versus strontium isotopic 87 86 composition ( Sr/ Sr) of CF and IP bulk rocks. (c) The ratio of Al2O3/K2O versus strontium isotopic composition (87Sr/86Sr) of CF and IP bulk rocks. Symbols are the same as defined in Figure 2 of the main text. Sources of bulk rock literature data are: (Albini et al., 1977; Beccaluva et al., 1991; Brocchini et al., 2001; Cannatelli et al., 2007; Capaldi et al., 1972; D'Antonio et al., 1999a; D'Antonio et al., 1999b; De Astis et al., 2004; de Vita et al., 1999; Di Girolamo et al., 1984; Di Vito et al., 2011; Ghiara et al., 1979; Mastrolorenzo and Pappalardo, 2006; Orsi et al., 1995; Orsi et al., 1992; Pappalardo et al., 2002; Piochi et al., 2005; Piochi et al., 2008; Scarpati et al., 1993; Tonarini et al., 2009; Tonarini et al., 2004; Turi et al., 1991; Villemant, 1988; Wohletz et al., 1995).

Supplementary Figure 2. Comparison MI and host mineral compositions. The plots show mineral compositions (Y-axis) versus the SiO2 concentrations of MI (X- axis). (a) Concentration of SiO2 in MI versus orthoclase contents of feldspar hosts. (b) Ratio of SiO2/CaO of MI versus orthoclase contents of sanidine hosts. The negative correlation of these two variables is consistent with crystallization of late stages of crystallization (Melluso et al., 2012). (c) Concentration of SiO2 in MI versus ferrosilite contents of clinopyroxene hosts. Note the positive correlation between the two variables is consistent with trapping of MI during fractional crystallization. Large symbols are MI and host data from this study and small symbols are MI and host data from the literature.

Supplementary Figure 3. Summary of various methods tested to correct compositions of anomalous MI. (a) Concentrations of CaO vs. Al2O3 for anomalous MI (red circle symbols). Trajectories in CaO vs. Al2O3 space that result if compositions of anomalous MI hosted in clinopyroxene are corrected by incorporating (dissolving) host material into the MI to account for overheating of the MI are shown by red solid and dashed lines. Trajectories to correct the compositions for PEC are shown by blue solid and dashed lines. The correction was based on regression (black line) of bulk rock data (small symbols). (b) Concentrations of SiO2 versus K2O showing the trajectory in SiO2 versus K2O space of corrected MI. Note that the correction for host dissolution results in a significant offset of these MI away from the bulk rock trends. (c) Concentrations of SiO2 versus FeOtot resulting from dissolution and PEC corrections. Note that MI corrected for PEC still do not align to the bulk rock trend. (d) Concentrations of SiO2 versus P2O5 resulting from dissolution and PEC corrections. Note that the correction predicts trends in the opposite direction from the bulk rock trend, especially for anomalous MI corrected for host dissolution (overheated MI).

Supplementary Figure 4. Concentration of K2O in MI versus size of MI hosted in sanidine. Note that in general the most K2O-rich MI are MI that were reheated in the lab, suggesting that the MI were heated to a temperature greater than the trapping temperature. Also, no apparent correlation between these two variables is noted for the unheated MI, i.e., MI that contained only glass as found.  3KRQROLWH a  7HSKUL  SKRQROLWH  2ZW

 7UDFK\WH  /DWLWH 7UDFK\EDVDOW 2. 

 BR IP 6KRVKRQLWH BR CF 1D  Xenoliths

 %DVDOW          

6L2ZW  b

 6U   6U 



     0J2ZW

 c

 6U   6U 



                

$O2.2 Suppl. Fig. 1  a



 2UPRO



      

6L2ZW  b 

 2UPRO



     

6L2&D2  c    

)VPRO         

6L2ZW Suppl. Fig. 2 a   33% 25%

30% 39% 30%

ZW 33% WRW ZW  30% 25%

2    )H2 39% $O 30% This study: Normal MI hosted in cpx c Anomalous   MI hosted in cpx           cpx hosts LLD from &D2ZW 6L2ZW rhyolite-MELTS  Literature:  Bulk rocks of CF b 33% 30% Bulk rocks of IP Xenoliths of CF  33% BR regression PEC trajectory

 25%

30% ZW Host dissolution   30% 2ZW

 trajectory

2 30%  .

3 

25% 39%  d 39%                        

6L2ZW 6L2ZW Suppl. Fig. 3  heated MI hosted in san unheated MI hosted in san 

 6L]H—P



    

.2ZW

Suppl. Fig. 4