Crystals in rhyolitic melts can nucleate multiple bubbles: an experimentally constrained model Quinn Johnson, James E Gardner, Wade Aubin, Patrick Sullivan, Fabian B Wadsworth, Edward W Llewellin, Tamara Carley, Halim Kusumaatmaja, Dork Sahagian Abstract Heterogeneous nucleation of bubbles on crystals in magma is thought to be an important driver of volcanic eruptions. Decompression experiments were carried out to elucidate the kinetics of H2O bubble nucleation on crystals of different sizes. Samples used were mixtures of hydrated rhyolitic melt and populations of augite or hornblende crystals of < 20 μm or 20–40 μm. We found that bubbles nucleated on surfaces of both types of crystals at supersaturation pressures ≤ 10 MPa, which is a small fraction of the supersaturation needed to homogeneously nucleate H2O bubbles in the melt. Very small crystals (< 5–10 μm) will nucleate only ~ 1 bubble, whereas on larger crystals, the number of bubbles that nucleate at a given supersaturation increases with an increased crystal surface area. In addition, the number of bubbles that nucleate on crystals of a given surface area increases with increasing supersaturation, indicating that the distance between nucleation sites decreases with increased supersaturation. We propose that the closer spacing of nucleation sites at higher supersaturation arises from the competition between bubble nucleation rate and the rate of H2O diffusion. Based on these results, we develop a modified heterogeneous nucleation rate equation that accounts for the control of crystal surface area on heterogeneous bubble nucleation and the resulting bubble number density generated. We provide a spreadsheet that allows users to calculate both total bubble number density and number densities per crystal size for a specified crystal-size population and different conditions (H2O content, temperature, and final pressure). Introduction Bubbles drive volcanic eruptions, and it is bubble-scale processes that underpin explosive eruptive behavior (Verhoogen ; Sparks ). Determining the rates at which bubbles nucleate is thus critical for understanding volcanism on Earth and other planets. Gas bubbles can nucleate either homogeneously within pure liquid or heterogeneously, where bubbles form on substrates such as crystals in magma (Toramaru ; Mourtada-Bonnefoi and Laporte , ; Gardner et al. ; Mangan and Sisson ; Iacono-Marziano et al. ; Gardner ; Larsen ; Cluzel et al. ; Cichy et al. ; Gardner ; Fiege and Cichy ; Le Gall and Pichavant ; Shea ; Plese et al. ; Cáceres et al. ; Gardner et al. ). Gardner et al. () proposed that during magma ascent, conditions may lead to bubbles nucleating via both mechanisms, leading to an overall system-scale bubble nucleation rate J, which is the sum of the heterogeneous and homogeneous contributions, JHOM and JHET: where Jo is the homogeneous nucleation prefactor, Ji is the pre-factor for the ith heterogeneous site (e.g., the ith crystal phase), σ is surface tension, kB is Boltzmann’s constant, T is temperature, Δp is the difference between the saturation pressure and the melt pressure, and αi is the heterogeneous adjustment factor for the ith phase (see Table for complete list of variables used in this study). The parameter βi was added to account for the number density of ith sites. Whereas homogeneous nucleation is reasonably well understood, with extensions beyond classical nucleation theory able to predict JHOM for rhyolitic magmas to reasonable accuracy (e.g., Gonnermann and Gardner ; Hajimirza et al. ; ), JHET is less well-constrained. Importantly, the efficiency of heterogeneous nucleation depends on the type of crystal present as expressed through αi, which has been approximated for only magnetite, hematite, clinopyroxene, and plagioclase (Hurwitz and Navon ; Gardner and Denis ; Larsen ). Numbers, sizes, and shapes of crystals should also impact heterogeneous nucleation (Schmelzer ; Schmelzer and Mahnke ; Baidakov and Boltachev ; Qian and Ma ; Hedges and Whitelam ), all of which are implicitly included in βi, but neither βi nor Ji have been calibrated, tested, or derived. In this study, we perform decompression experiments targeted to calibrate αi for clinopyroxene and hornblende, both common minerals in rhyolite magma. We find that both are very efficient sites for nucleating H2O bubbles, allowing nucleation at supersaturations (Δp) as low as 5 MPa. We also find that the number of bubbles nucleated on crystals larger than ~5 μm is controlled by the available surface area and Δp. Based on these results, we modify the formulation of the heterogeneous term in Eq. by combining βi and Ji into a single heterogeneous rate term, Jc, that captures the microphysics of heterogeneous nucleation, aligning better with nucleation theory (e.g., Blander and Katz ). Methods Experiments consisted of mixtures of hydrated silicate glass and crystals. The glass was made using 3-mm-diameter cores drilled from a high-silica rhyolitic obsidian (Table ). Each core, and enough distilled H2O to saturate it at experimental run conditions, was placed inside a gold capsule that had been welded shut at one end. The other end was then welded, and the capsule was heated to 110 °C for 5 min to ensure there were no leaks in the welds. A sealed capsule and Ni filler rod were placed inside a cold-seal pressure vessel, which was then pressurized by water to 150 MPa and heated to 850 °C for 5 days (time sufficient for diffusive hydration of the core; Gardner et al. ). The hydrated rhyolite glass was then crushed and ground to < 45 μm with an agate mortar and pestle. Crystals used were either augite (NMNH R3035) or hornblende (NMNH 117329-1) obtained from the Smithsonian Institution. Mineral compositions are given in Table . Each crystal was crushed in a steel mortar, and the fragments were wet sieved to be between 4.5 φ and 5.5 φ, referred to as the 20–45 μm size fraction. A separate sample of < 5.5 φ fragments (≤ 20 μm) was also used. An experiment consisted of ~40 mg of hydrated glass powder mixed with ~ 0.6 (± 0.1) mg of crystals, which approximates 1 vol% crystal contents. The mixture was poured into a 4-mm-diameter gold capsule that had been welded shut at one end. The other end was left open so that the sample could equilibrate with the pressurizing H2O to facilitate sintering of the particles (Gardner et al. , ; Aubin and Gardner ). The capsule was placed inside the sample cup at one end of an Inconel rod, which was then inserted into a pressure vessel equipped with a rapid quench extension (Gardner ). The sample was held in the water-cooled chamber of the vessel while the pressure vessel equilibrated to experimental conditions of 875 °C and 150 MPa. Once equilibrated, the sample was raised into the hot zone using an external magnet and held at those conditions for 1 h while the mixture sintered. After 1 h, pressure was increased rapidly to 200 MPa and held for 10 min in order to resorb as many bubbles trapped during sintering as possible (Gardner et al. , ). The higher pressure raises the solubility of H2O in the melt resulting in some H2O held in bubbles formed by sintering to dissolve into the melt and thereby shrinking (or completely resorbing) sintering bubbles. After 10 min, there are far fewer bubbles still present in the samples than before compression. Seven samples were quenched after the 10 min by rapidly lowering them back into the water-cooled jacket, where they cooled at ~150 °C per second (Gardner et al. ). Eight other samples were decompressed to pressures between 150 and 185 MPa in ~ 1–2 s and held at the new lower pressure for 5 min before being quenched. Note that part of one sample (Q-90) that had been quenched at 200 MPa was used as starting material for one decompression experiment (Q-108) that followed all other protocols. All samples were prepared as thick sections for analysis. The seven samples quenched at 200 MPa were doubly polished and analyzed for dissolved H2O contents using Fourier Transform Infrared (FTIR) spectroscopy (Table ). H2O contents were determined from measured peak absorbances at 5250 cm−1 and 4500 cm−1 wavenumbers using the model of Zhang et al (). Sample thickness was measured at each analyzed spot by focusing through the sample using a petrographic microscope that is equipped with a Heidenhain linear encoder that measures the vertical motion of the microscope stage (Gardner et al. ; Aubin and Gardner ). H2O contents listed in Table are averages of 4–8 measurements, and uncertainties are the standard error on those averages. All decompressed samples were analyzed for crystal and bubble numbers and sizes (Table ). The three dimensions of thirty (30) randomly selected crystals in each sample were measured using the petrographic microscope following the methods of Befus et al (), and then, the number of bubbles that were clearly attached to each crystal was counted. Because augite and hornblende are reasonably transparent, bubbles attached to all sides of the crystals could be counted (Fig. ). Another fifty (50) crystals were randomly selected in each sample that used 20–45 μm crystals, and bubbles were counted on those crystals while noting whether each bubble appeared to nucleate on a flat crystal face or at the corner or edge of the crystal. Results Starting conditions Dissolved H2O contents in samples quenched after equilibrating at 200 MPa are, on average, 5.76 ± 0.16 wt% (Table ). This H2O concentration equates to a saturation pressure of ~ 195 MPa, based on the solubility model of Liu et al (). We thus assume that all other samples had saturated at 195 MPa before pressure was lowered, except for Q-108, which used part of Q-90 as starting material. Q-90 had 5.48 wt% dissolved H2O, equal to a saturation pressure of 180 MPa. Careful examination of the augite and hornblende crystals indicates that no significant change in shapes from either melting or crystallization occurred during the experiments. Some crystals in the samples quenched at 200 MPa have bubbles attached to them, formed during sintering (Blandon et al. ). Thirty crystals in two samples (one containing augite and one containing hornblende) were selected, and the number of these sintering-derived bubbles was counted, and averages 1 ± 1 bubbles per crystal (Table ). We assume that the same number of sintering-derived bubbles formed in all samples, and thus, for the decompression experiments, we subtracted 1 from the number of bubbles counted on each crystal (unless no bubbles were attached). Given the large number of bubbles on the surface of crystals, we take the resulting error to be small in the subsequent analysis. Small magnetite crystals (≤ 5 μm in size) grew in the melt during sintering and compression (they were not present after the initial 5-day hydration). Such crystals also formed in similar experiments reported in Aubin and Gardner (). Their presence is unavoidable, and we ignore them for this study. Decompression experiments All samples that were decompressed have an outer rind, a few 10 s of microns thick, that contains numerous small bubbles. Such rinds are ubiquitous in experimental decompressions and result from bubbles nucleating at the contact between the sample and metal capsule (e.g., Mourtada-Bonnefoi and Laporte ; Mangan and Sisson ; Gardner ; Aubin and Gardner ). We avoided those rinds and focused on crystals located in the interior of the samples (Fig. ). As seen in Fig. , there are bubbles in the interior of all samples that did not nucleate on the augite or hornblende crystals. Some of these bubbles are relicts from the sintering process that did not completely resorb during the compression step. Others nucleated on the small magnetite crystals that formed during the sintering process (Aubin and Gardner ). Bubbles readily nucleate on magnetite at small degrees of supersaturation (Hurwitz and Navon 1995; Aubin and Gardner ), and thus, their presence is unavoidable. For this study, we ignore all bubbles that did not nucleate on augite or hornblende crystals. Four decompressions used augite crystals (Table ). Three used the 20–45 μm crystal-size fraction; those crystals have mutually perpendicular principal axes of average length 39.9(± 22), 15.0(± 7.7), and 10.0(± 4.8) μm (e.g., Fig. a, b). The fourth used the < 20 μm sized fraction with lengths 5.4(± 2.0), 2.6(± 0.7), and 2.4(± 0.9) μm. Four decompressions used hornblende crystals (Table ). Three used the 20–45 μm size fraction of hornblende, with average lengths of 36.8(± 16.4), 16.1(± 7.1), and 11.1(± 4.7) μm (Fig. c, d). The fourth used smaller sized hornblende with lengths 14.5(± 7.9), 5.4(± 2.5), and 5.0(± 2.9) μm. The nearly instantaneous pressure drops generated supersaturation pressures (Δp) of 5 to 45 MPa, equal to the difference between saturation pressure (based on initial H2O contents) and final pressure given in Table . Regardless of Δp, all bubbles that had nucleated on augite and hornblende are found to be 1–6 μm in diameter (Fig. ). In samples that used the 20–45 μm size fractions, the proportion of crystals that seeded bubbles increased as Δp increased. At Δp = 5 to 10 MPa, bubbles nucleated on ~50% of the augite and amphibole crystals, whereas bubbles nucleated on 87–93% of crystals at Δp = 20 to 25 MPa, and bubbles nucleated on every crystal at Δp = 40 to 45 MPa. The number of bubbles nucleated on each crystal increases as Δp increases (e.g., Fig. a, b). At Δp = 5 to 10 MPa, 64–94% of augite and hornblende crystals seeded 1–2 bubbles (in excess of those formed during sintering), with up to seven bubbles on a few of the crystals. At Δp = 20 to 25 MPa, an average of 5–10 bubbles nucleated on crystals, with up to 28 on some (Fig. a). At Δp = 40 to 45 MPa, 2 to 72 bubbles nucleated on crystals, with more than 60% of them seeding 20 or more (Fig. b, d). Interestingly, similar values of Δp result in similar numbers of bubbles on augite and hornblende (Table ). The locations of 4037 bubbles were analyzed, and 97.5% of them appear on flat surfaces as opposed to being on corners or edges, regardless of Δp or crystal type (Fig. ). At a given Δp, more bubbles nucleated on larger crystals than on smaller ones (Fig. c, d). To investigate this relationship further, we calculated the surface areas of crystals in decompressions that used the 20–45 μm size fraction by assuming all are prismatic cuboids (Fig. ). We find that augite has surface areas of 167–9248 μm2 and hornblende are 254–7629 μm2. There is a strong relationship between the number of bubbles on a crystal and the surface area of that crystal at any given Δp (Fig. ). At similar values of Δp, the number of bubbles per surface area is nearly the same on hornblende and augite crystals. By calculating the number of bubbles per surface area of each crystal, we find that the average number of bubbles per surface area increases with increased Δp (Fig. ). Discussion Overall, bubble nucleation on augite and hornblende is almost identical at a given Δp in terms of the number of bubbles nucleated, the number of bubbles per crystal surface area, and the proportions of crystals that seeded bubbles. We thus assume in the discussion below that the kinetics of heterogeneous bubble nucleation on augite and hornblende in a rhyolitic melt are the same within experimental uncertainty. Efficiency of augite and hornblende for bubble nucleation Modeling heterogeneous bubbles nucleation requires knowing the value of αi (e.g., Eq. ). It is very difficult to accurately measure the wetting angles of bubble on crystals in our experiments, making it impossible to determine αi from the wetting angles of bubble on crystals (e.g., Hurwitz and Navon ). To first order, however, αi can be approximated from the ratio of the minimum supersaturation pressures required for homogeneous or heterogeneous nucleation in the same melt from the following, where ΔpHOM and ΔpHET are the minimum supersaturations for homogeneous and heterogeneous nucleation, respectively (Hurwitz and Navon ; Hajimirza et al. ). In this study, bubbles nucleated on hornblende at Δp = 5 MPa and on augite at 10 MPa, which suggests that ΔpHET ≤ 5 MPa for both. The high-silica rhyolitic melt used in this study has been used in other studies of homogeneous bubble nucleation (Gardner and Ketcham ; Gonnermann and Gardner ; Hajimirza et al. ). Those found that when the melt contained ~ 5.5–6.0 wt% dissolved H2O, bubbles do not nucleate homogeneously at Δp = 100–115 MPa but do at Δp ≥ 125 MPa, indicating that ΔpHOM = 120 ± 5 MPa. From these supersaturation pressures, we estimate that αi ≤ 0.0017 ± 0.0002 for both hornblende and augite (Table ). In other studies, bubbles were found to nucleate on magnetite in hydrous rhyolitic melts at Δp ≤ 5 MPa (Hurwitz and Navon ; Gardner and Denis ; Aubin and Gardner ). Augite and hornblende are thus similarly efficient as magnetite for heterogeneous bubble nucleation. In contrast, plagioclase does not seed bubbles in rhyolitic melt (Hurwitz and Navon ; Gardner and Denis ), and hence, αi = 1. In contrast to those results in rhyolitic melts, Plese et al () found that bubbles readily nucleate on plagioclase in andesitic melts, and Larsen () found that Δp must be ~ 50 MPa (αi ~ 0.17) for bubbles to nucleate on pyroxene in potassic phonolitic melts. More studies of heterogeneous bubble nucleation are thus needed to fully understand the relationship of α with crystal type and melt composition. Gardner and Denis () found that bubbles nucleated on the tips of hematite rods and along the lengths of those rods at very different supersaturations suggesting that αi can vary with crystal structure or the different geometry at the tips versus along the sides (e.g., Schmelzer ; Schmelzer and Mahnke ; Baidakov and Boltachev ; Qian and Ma ; Hedges and Whitelam ). In this study, bubbles nucleated at very low values of Δp mainly on flat surfaces (97.5% of the bubbles) but also on corners and edges of both amphibole and augite. Thus, if there are any differences in nucleation kinetics with amphibole or augite structure or surface geometry, it must occur at extremely small values of Δp. Spacing between nucleation sites on augite and hornblende The number of bubbles nucleated on small augite and hornblende approaches 1 per crystal as surface area approaches 200 μm2, even at Δp =45 MPa (Fig. ). Aubin and Gardner () also found that at high values of Δp only one bubble nucleated on magnetite crystals that were ~ 1–5 μm in size (surface areas = 6–150 μm2). Crystals larger than ~ 200 μm2, however, can seed more than one bubble, and the number increases with surface area and higher Δp (Fig. ). Aubin and Gardner () found that the number of bubbles that nucleated on large magnetite also increased as Δp increased, but because magnetite is opaque, not all sides could be examined, and so, we cannot determine the complete number of bubbles per surface area. The average number of bubbles per surface area increases with increased Δp (Fig. ), such that at Δp = 10 MPa, there is, on average, 1 bubble per ~2500 μm2, but at Δp = 45 MPa, there is ~1 bubble per ~90 μm2. If we assume that every bubble represents a nucleation site that was activated by a given value of Δp, then we can calculate the average spacing between sites from the square root of the surface area. We find that the average spacing decreased from ~30 μm at Δp = 10 MPa to ~ 9 μm at Δp = 45 MPa (Fig. ). The distance between nucleation sites is thus on average shorter at higher values of Δp (Fig. ). We interpret this shrinking distance between nucleation sites to reflect the tradeoff between the rate of nucleation and the rate of H2O diffusion into bubbles already nucleated. As Δp increases, the free-energy barrier to heterogeneous bubble nucleation decreases rapidly, leading to a strong increase in JHET (Eq. ). As such, at greater Δp, diffusion of dissolved H2O into nearby bubbles is too slow to prevent further nucleation, thus leading to a greater number of bubbles nucleated per unit area. Modeling heterogeneous bubble nucleation on crystal populations Gardner et al () proposed that JHET should be modified to include the number density of sites available for bubbles to nucleate (i.e., the βi term in Eq. ). That correction assumed that there is one nucleation site per crystal, which our experiments show holds only for crystals smaller than ~ 5 μm. On larger crystals, the number of nucleation sites increases with surface area and depends on Δp (Figs. and ). We now use these results to reformulate JHET to include that dependence on crystal surface area. Given that we are considering nucleation as a function of surface area, we first consider a population of nucleated bubbles in terms of area density N′ (instead of volume density N). The characteristic time taken for a bubble to nucleate on a given surface area can be predicted as where JHET′ is the nucleation rate per unit area per unit time, and Anuc is the area of the crystal per bubble, which is the inverse of the area density of bubbles: Anuc=1/N′. Expressing nucleation rates on a per unit area basis is consistent with theoretical treatments of heterogeneous nucleation processes (Blander and Katz ), and it is straightforward to convert between JHET′ (per area, Eq. ) and JHET (per volume, Eq. ), as we will show later. As we note above, observations of very small crystals show that the surface area required to nucleate a single bubble is much smaller than the value of Anuc measured from larger crystals that host multiple bubbles. This implies that nucleation rate is not controlled exclusively by the availability of nucleation sites on the crystal surface—some other process is limiting nucleation on larger crystals. We infer that, once a bubble nucleates on a crystal surface, diffusion of H2O into the bubble reduces supersaturation locally and suppresses the formation of new bubbles within an area that grows with time. This defines a characteristic diffusion timescale. where D is the diffusivity of H2O. This is the characteristic time for diffusion into a bubble to deplete H2O from an area of crystal surface around it of crystal-size Anuc. We propose that the area density of nucleation events on a crystal is controlled by the balance between the characteristic time to nucleate a particular number of bubbles per unit area tnuc, and the characteristic time for diffusion to inhibit bubble nucleation tdiff. Note that this definition of tdiff differs from that used in Toramaru and Miwa (), who measure the diffusive timescale associated with the growth of bubbles whereas we present the timescale describing the growth of the diffusive exsolution halo, inside which nucleation is suppressed. No more nucleation will occur when the entire crystal surface has been covered by these exclusion haloes. For a given area density of bubbles, we now have two measures of when nucleation will be completed. By setting tnuc=tdiff, we thus obtain the expression which gives the nucleation area density in terms of the per unit area nucleation rate and diffusivity. We can rewrite the formulation for JHET′ to give where Jc′ is the pre-exponential factor. This formulation is similar to the formulation of JHET in Eq. , except that it is on a per-area basis rather than a per-volume basis, and it does not include the term βi (the effects that βi encoded will be discussed below). This definition of JHET′ leads to an expression for the nucleation density on a per-area basis. Importantly, because we can calculate the appropriate values for diffusivity (D=10-10.6m2s-1 for rhyolite melt with 5 wt% H2O; Zhang et al. ) and surface tension given by the Tolman-corrected function σ(Δp) from Hajimirza et al. (), and we have determined α = 0.0017 from our experiments, this leaves Jc′ as a fit parameter to compare solutions to Eq. 7 to our experimental dataset (Fig. ). We find that a value of Jc′≈1010m-2 matches the experimental results well, with the caveat that this holds only for the experimental conditions. In the formulation above, N′ is the number of bubbles per area of crystal. As shown in Fig. , the number of bubbles increases approximately linearly with crystal area; this linearity does not hold for very small crystals, however, which tend to host a single bubble even when the total surface area of a crystal (A) is much smaller than Anuc. This is readily explained within the diffusion-limited mechanism that we propose for larger crystals. Self-evidently, the diffusive shut-off mechanism can only apply after at least one bubble has formed on the surface of a crystal. Therefore, the first bubble to form on any crystal (which is the only bubble to form on crystals with A≪Anuc) is not subject to the diffusion limit; this manifests as a higher apparent N′ for these small crystals. In order to capture the transition in regimes from one bubble per crystal to multiple bubbles per crystal, we introduce a correction factor χ to give Ncorr′=JHET′χ/D, where χ is given by This parameter is a correction to the predicted bubble number density to capture the transition from one bubble per crystal regime at low A to multiple bubbles per crystal regime at high A. This allows us to formulate an expression for the expected number of bubbles on a crystal N as a function of the crystal surface area A. The increase in χ at low A ensures that at least one bubble will form on every crystal. Noting that Anuc=D/JHET′ (from setting tnuc=tdiff in Eqs. and ), and that N=N′A, we find In the limit of A≪Anuc, this reduces to N=1, and in the limit of A≫Anuc, this reduces to N=N′A, as required. We note that this approach captures the effects that Gardner et al () originally cast as the parameter βi (see Eq. ). By using our previously determined value for Jc′ (1010m-2), we can use Eq. to forward model the number of bubbles per crystal area (Fig. ). Solutions for different values of Δp match the experimental data well. So far, we have determined heterogeneous nucleation in terms of area density. From this, we can determine the volumetric number density of bubbles, Nb, nucleated in magma containing crystals in number density Nc. For monodisperse crystals with identical surface area A, we have More generally, for crystals of different sizes, we have where F(A) is the crystal surface area distribution (i.e., the number of crystals with area A per unit volume in units of m−3) such that Nc=∫0∞FAdA. Therefore, FA can be explicitly found from a crystal-size distribution. Note that in our analysis and model, crystal-size distribution and bubble number distribution refer to sizes or numbers on a per-volume basis (with units of m−3) and should not be confused with population density distributions (m−4). The number of bubbles nucleated per unit volume in the system can be calculated from the supersaturation pressure and crystal-size distribution from either Eqs. or , depending on the size distribution of crystals. Using the characteristic nucleation time from Eq. , we can calculate JHET (for an individual crystal type) by comparing the number of bubbles that nucleate on the crystal with the nucleation time and volume of melt surrounding the crystal. Here, it becomes more convenient to express the nucleation time as tnuc=1/JHET′D by eliminating Anuc from Eqs. and , which results in JHET being given by As with Eq. , integrating over the crystal population allows the nucleation rate for the entire system to be retrieved, giving which takes the form of the heterogeneous nucleation rate equation in Eq. , allowing us to identify the relationships between our per unit area analysis of nucleation on large crystals and the volumetric nucleation rate. At present, this approach assumes that there is nucleation on every crystal. Our study demonstrates that in practice, diffusion of volatiles between crystals will reduce this effective rate at sufficiently low values of Δp. Calculation of bubble number density for a known crystal population We provide a spreadsheet that allows users to enter crystalsize data and known parameters (temperature, pressures, dissolved H2O contents) to estimate both the number density of bubbles nucleated on each crystal-size and the total number density of bubbles (see ). The number of crystals of each size class per unit volume can be calculated from a known distribution of crystals using the population density distribution of sizes n′ (not to be equated with Nc used earlier). Here, n′=lCexp(ml), where l is the crystal length and m and C are constants that capture the slope and intercept of the population distribution. FA is then computed as the area of crystal per volume of melt as FA=n′A, where area is calculated. In the spreadsheet, we assume crystals are cubic shaped. As an example, we use the crystal-size distribution of clinopyroxene microlites generated in decompression experiments using Inyo pyroclastic rhyolite (Castro and Gardner ). For our purposes, we extend the size range to 50 μm but maintain the overall distribution which is skewed toward greater numbers of small crystals (Fig. ). We assume an initial H2O content of 5 wt%, T = 875 °C, and α = 0.0017. We calculated bubble number density distributions for two different supersaturations (Fig. ). As expected, the number densities of bubbles nucleated (i.e., bubbles per unit volume of magma) on crystals smaller than ~ 5 μm equal those of the crystals, but on larger crystals, bubble number densities far exceed crystal number densities. The excess of number of bubbles on larger crystals increases with crystal size (area) and Δp. The peak in bubble number densities occurs on ~20–25 μm crystals, but that results from the crystal-size distribution being skewed toward small sizes. Overall, the total number density of bubbles nucleated at Δp = 15 MPa is 1013.2 m−3 and 1013.6 m−3 at Δp = 50 MPa. In our examples, Nb exceeds Nc by a factor of three at Δp = 15 MPa and a factor of seven at Δp = 50 MPa. This illustrates that the number density of heterogeneously nucleated bubbles depends strongly on the crystal population because larger crystals provide greater numbers of nucleation sites. Indeed, in our example, the number density of bubbles on 40-μm crystals equals the number density on 8-μm crystals, despite there being about ten times more smaller crystals. We caution that our model is built by assuming that nucleation occurs on every crystal. In practice, this is reasonable for magnetite, augite, and hornblende at values of Δp ≥ 15–20 MPa (Aubin and Gardner ; this study). Any prediction of Nb at lower values of Δp would likely be overestimated. Our examples also assume α = 0.0017, which is reasonable for magnetite, augite, and hornblende. Predicted values of Nb will likely differ for other crystal types. Finally, our calculations assumed Jc=1010m-2, which is valid for augite and hornblende over the range of Δp investigated (Fig. ). Regardless of such caveats, our modeling shows that heterogeneous bubble nucleation depends on the full characteristics of the crystal population and not merely on whether a particular mineral type exists. Conclusions Decompression experiments were carried out using hydrous rhyolitic melts seeded with populations of augite or hornblende crystals. The number of bubbles that nucleated on crystals depended strongly on the degree of H2O supersaturation in the melt generated by the decompression (Δp) and on the surface area of the crystals present. Crystals larger than ~ 5 μm (those with more surface area) allow more bubbles to nucleate, with the average number of bubbles per surface area increasing with Δp. Smaller crystals seed only one bubble, regardless of Δp. The number of nucleation sites on large crystals increases with Δp because the rate of heterogeneous nucleation increases faster than the rate of H2O diffusion into bubbles already nucleated. There are no significant differences in the behavior of bubbles nucleating on augite and hornblende, and the efficiency of heterogeneous nucleation on those crystals equals that found on magnetite by previous studies. These results enabled us to reformulate the equation for calculating the rate of heterogeneous bubble nucleation to include not only the heterogeneous adjustment factor αi but also a modified pre-exponential factor that accounts for the characteristics of the crystal population on which the bubbles nucleate. Future studies can expand this analysis for other crystal types and melt compositions. Footnotes