Gas-driven filter pressing is the process of melt expulsion from a volatile-saturated crystal mush, induced by the buildup and subsequent release of gas pressure. Filter pressing is inferred to play a major role in magma fractionation at shallow depths (<10 km) by moving melt and gas relative to the solid, crystalline framework. However, the magmatic conditions at which this process operates remain poorly constrained. We present novel experimental data that illustrate how the crystal content of the mush affects the ability of gas-driven filter pressing to segregate melt. Hydrous haplogranite (2.1 wt% water in the melt) and dacite (4.2 wt% water in the melt) crystal mushes, with a wide range of crystallinities (34–80 vol% crystals), were investigated using in-situ, high-temperature (500–800 °C) synchrotron X-ray tomographic microscopy with high spatial (3 μm/pixel) and temporal resolution (∼8 s per three-dimensional data set). Our experimental results show that gas-driven filter pressing operates only below the maximum packing of bubbles and crystals (∼74 vol%). Above this threshold, the mush tends to fracture and gas escapes via fractures. Therefore, the efficiency of gas-driven filter pressing is promoted close to the percolation threshold and in situations where a mush inflates slowly relative to build-up of pressure and expulsion of melt. Such observations offer a likely explanation for the production of eruptible, crystal-poor magmas within Earth’s crust.


Magmatic differentiation involves the physical separation of crystals from their coexisting melts. In shallow magma reservoirs, relatively slow, yet poorly constrained, compaction processes at high crystallinities (≥70 vol% crystals; Jackson et al., 2003) and hindered settling at intermediate crystallinities (40–50 vol% crystals; Bachmann and Bergantz, 2004) can be enhanced by the concentration of volatiles in the melt phase and their subsequent exsolution (Sisson and Bacon, 1999). Volatile exsolution at low pressures causes the magma to expand, while the high viscosity of the crystallizing magma (>105 Pa·s) impedes the bulk inflation of the system. In this scenario, gradients in crystallinity, vesiculation, and pressure therefore drive the melt toward regions of lower crystallinity and pressure. SiO2-rich melts are saturated with 6–8 wt% H2O at the depths of <10 km (Hui et al., 2009) typical of most silicic magma reservoirs, suggesting that this process may be ubiquitous. Gas-driven filter pressing may drive segregation of compositionally variable, crystal-poor melts of the type that form large ignimbrite deposits (Lipman et al., 1966). The goal of the present study is to quantify the influence of the crystal fraction (ϕ) on melt permeability in order to define the magmatic conditions where gas-driven filter pressing is an efficient mechanism for melt extraction and generation of eruptible, crystal-poor silicic magmas.


To capture simulated gas-driven filter pressing, high-temperature (T = 500–800 °C) experiments were conducted on a suite of pre-synthesized, highly differentiated, volatile- and crystal-poor melts (haplogranite, H5; 2.1 wt% H2O in the glass, ϕ = 0.34 and 0.47 corundum crystals) and less evolved, volatile- and crystal-rich melts (dacite, F; 4.2 wt% H2O in the glass, ϕ = 0.5, 0.6, 0.7, and 0.8 quartz crystals) using high spatial (3 μm/pixel) and temporal resolution (∼8 s per three-dimensional [3-D] data set) of synchrotron X-ray tomographic microscopy at the Tomographic Microscopy and Coherent Radiology Experiments (TOMCAT) beamline (Swiss Light Source, PSI, Villigen, Switzerland), coupled to a laser-based heating system (Fife et al., 2012). Crystal-free samples of both compositions were used as a benchmark during experiments. The starting materials have bubble volume fractions (β) ≤0.01 and some heterogeneity in the crystal distribution (mean size of 68 μm; Fig. 1); however, neither crystallization nor melting of crystals occurred during heating (i.e., ϕ remained constant). We did not simulate crystallization-driven gas exsolution; rather the different ϕ values bracketed the crystallinities occurring in natural gas-saturated mushes. The limited attenuation contrast between crystals and melt was maximized by edge enhancement and post-acquisition phase retrieval. Technical details are reported in the GSA Data Repository1.

The haplogranite and dacite have the same initial melt viscosity (Δηmelt < 1.2 Pa·s). Preferential bubble nucleation on the crystal phases is not expected in either system, as SiO2-rich melts are the wetting phase (Laporte, 1994; Hurwitz and Navon, 1994). Although both systems were not expected to crystallize during the experiments, chemical differences between the samples (e.g., SiO2 content, initial H2O content) may affect bubble nucleation and growth during T increase to 800 °C. We do not assume a priori that both systems have the same physicochemical behavior; rather we explore which allows gas-driven filter pressing.

Sequential 3-D images provided a 4-D (3-D plus time) record of bubble growth and microstructure evolution for each T and ϕ as the samples were heated stepwise (25 °C steps with a heating rate of 2 °C/s) from 475 °C (below the glass transition) to 800 °C, with conditions maintained for 3.5 min at each T. Although the samples from this study were explored below the solidus at most T values, the supercooled silicic matrix behaved rheologically as a melt under experimental conditions. Real-time visual inspection showed negligible bubble growth by the end of each step, but samples did not achieve textural equilibrium at each T. However, gas-driven filter pressing is a process driven by rapid crystallization and vesiculation (Sisson and Bacon, 1999), and therefore operates during textural and thermal disequilibrium.

Karl Fischer titration (KFT; Behrens et al., 1996) was used to quantify H2O outgassing over the same T-time path for samples of the same volume to those used in the X-ray tomographic microscopy experiments.


Time-integrated textural analysis (see the Data Repository) reveals that gas bubbles nucleate and undergo diffusion-limited growth homogeneously throughout crystal-free samples. Fractures were only generated during vesiculation. No formation or healing of fractures was observed during cooling after experiments.

In the haplogranite crystal-bearing samples, there was no evidence for gas-driven filter pressing. Bubbles increased their volume with minimal coalescence, and formed a polygonal network (Figs. 1A–1D) similar to that found in natural felsic frothy pumices. Despite the initial intercrystalline porosity, bubble growth preferentially occurred in melt-rich regions (Fig. 1B). Within the crystal-rich regions, bubble distribution was homogenous. Bubble growth reduced the volume of the interstitial melt (local β = 0.9) and generated peripheral compacted crystal clusters (Figs. 1C and 1D). At T ≥ 550 °C major fractures developed (Figs. 1C and 1D, arrows), with smaller fractures up to 200 μm in length connecting inflated gas-rich regions (Fig. 1D). Fractures were arranged at high angles (70° to 90°) relative to the vertical sample axis along which expansion occurred. The fractures also radiated out from the inflating gas-rich regions, passing through both crystals and residual melt (Figs. 1C and 1D, arrows).

In contrast, the dacite samples showed gas-driven filter pressing. Individual bubbles were generally much larger than those in the haplogranite samples (Figs. 1E–1H). Over the entire T range, at ϕ ≤ 0.5, bubbles formed and grew by extensive coalescence (predominantly through melt-film attenuation) and no fractures were observed. At ϕ = 0.6–0.7, bubbles deformed around crystals during growth (Fig. 1H), and melt became concentrated into narrow (20–80 μm wide) channels within the crystal framework due to the pressure exerted by gas bubbles (Fig. 1H, black arrows from stretched bubbles). At T ≥ 675 °C, curved fractures formed between large bubbles in the melt phase, and jagged fractures were found in the crystal-rich regions (Fig. 1H, white arrows). At ϕ = 0.8, no significant vesiculation was observed at any T (see insets in Fig. 2).

From the observed behaviors we define the “ductile regime”, when the sample undergoes inflation during vesiculation, and the “brittle regime”, when the sample largely fractures during vesiculation (with or without inflation). The mechanical evolution can be described as a function of ϕ and β. At ϕ = 0.5–0.7 (the target crystallinities of this study), the ductile to brittle transition occurs at a residual melt fraction (μ = 1 − ϕ − β) of ∼0.25 (Fig. 2).


KFT analysis performed on the dacite samples showing gas-driven filter pressing supports the microstructural observations. Accelerated H2O extraction is favored at higher T and with decreasing ϕ (Fig. 3A). H2O extraction begins between 500 °C (ϕ = 0.7) and 625 °C (ϕ = 0) (Fig. 3A). At high ϕ, low permeability is maintained, whereas bubble coalescence allows gas loss at ϕ ≤ 0.5. At ϕ = 0.6–0.7, H2O extraction occurs across a restricted T range (≤75 °C), and, after the onset of brittle behavior, H2O extraction is below the limit of detection (∼0.02 wt%) at all T values. The onset of brittle behavior is accompanied by an increase in the KFT uncertainty, which suggests that fracture-enhanced permeability permits continuous low-volume “silent” emission of gas (see the Data Repository). At ϕ = 0.8, no H2O extraction is detected across the entire T range (Fig. 3A), meaning that the exsolved H2O must remain trapped in a non-permeable bubble network and/or be released via low-volume “silent” emission. These two processes may operate simultaneously.


Gas-driven filter pressing appears to be most efficient in crystal mushes (0.5 ≤ ϕ ≤ 0.7), with a minimum of ∼3 wt% H2O dissolved in the melt (Fig. 3B), and when bulk sample expansion occurs without fracturing or the development of gas permeable networks. In addition, filter pressing appears to be efficient over a limited window of crystallinity (0.6 ≤ ϕ ≤ 0.7) and requires μ > 0.25 (Fig. 2), close to the percolation threshold (μ = 0.22–0.29) or maximum packing fraction (ϕmax = 0.66–0.74 for monodisperse suspensions) (Saar and Manga, 2002). When ϕ ≤ 0.5, permeability via bubble coalescence prevents gas-driven filter pressing. At ϕ = 0.6–0.7, bubble growth drives filter pressing, until μ falls below the percolation threshold and exsolution alone drives pore-pressure increase until brittle failure occurs.

A wide range of igneous rocks attain the critical packing density of bubbles and crystals (ϕmax ∼0.65–0.75 in basalts [Marsh, 1981]; μ ∼0.3 in granites [Wickham, 1987]), which provides the final snapshot of a jammed system below the minimum volumetric proportion of melt to enable flow.

Melt viscosity (ηmelt), which should be the same at each T in all samples, evolves as a function of T and the residual dissolved H2O (Giordano et al., 2008), and strongly controls the effectiveness of gas-driven filter pressing. Bulk η of the sample will also depend on the local ϕ and β (Pistone et al., 2012).

As H2O exsolves, ηmelt increases slowly while H2O remains >2 wt%, and more rapidly at <2 wt% (Giordano et al., 2008). Where bubble coalescence is the dominant mechanism (ϕ ≤ 0.5), the increase of ηmelt due to H2O loss is more important than its decrease due to higher T (Fig. 3B). As is seen here, the same ηmelt is expected to occur at ϕ ≥ 0.6; however, the bulk H2O (i.e., dissolved H2O in the melt and exsolved gas bubbles) in the system remains in excess of 2 wt% due to the incapacity of the system to outgas in the presence of a continuous crystal network. This observation confirms that gas-driven filter pressing has maximum efficiency at 0.6 ≤ ϕ ≤ 0.7. After the brittle onset, samples enter the Mohr-Coulomb regime wherein ηmelt is meaningless (Fig. 3B).

The behavior of the dacite and haplogranite samples (the latter brittle at ϕ ≤ 0.47 due to the low H2O dissolved in the melt) suggests that, at equivalent rates of volatile exsolution as those simulated in our experiments (i.e., relatively fast T-time paths simulating rapid crystallization and vesiculation in natural systems), gas-driven filter pressing might only be effective if the residual melt achieves a sufficiently low ηmelt to prevent fracturing of the mush during gas exsolution, while maintaining a sufficiently high ηmelt to allow gas pressure to build up and expel melt from the crystal framework. Based on our results, the minimum H2O content in the silicic melt that allows gas-driven filter pressing to be effective in crystal mushes (ϕ ≥ 0.5) is ∼3 wt% (Fig. 3B). In addition, elevated pressure (<1.5 GPa) leads to a reduction in ηmelt of about two orders of magnitude (e.g., Pistone et al., 2012), which may further promote the efficiency of gas-driven filter pressing. Overall, hydrous dacite systems probably represent the optimal conditions of efficient gas-driven filter pressing to promote melt segregation from shallow plutonic mushes (<10 km).

To assess gas-driven filter pressing as a mechanism of melt extraction from shallow crystal mushes, we need to constrain the operating window controlled by melt permeability (κ) and crystal mush expansion rates that permit inflation without fracture (i.e., fracturing determines the viscous death of the system; Pistone et al., 2013). Extraction of silicic melts from a mush depends on κ, which is a function of melt fraction (μ) and crystal size (radius, r) (McKenzie, 1984): 

where Α = 50 for μ > 0.1 and r < 1 mm (Jackson et al., 2003). Excluding systems with ϕ ≤ 0.5 due to large gas permeability, the μ considered ranges from 0.25 (minimum percolation threshold) to 0.4. If r = 68 μm, κ ranges from 2.7 × 10−12 m2 at μ = 0.25 to 7.05 × 10−12 m2 at μ = 0.4 in our experiments. The average velocity of melt percolation (v, positive upward within the mush) can then be estimated using Darcy’s law: 

To maintain a gas pressure gradient (∇P) sufficient to expel melt, inflation must be slower than crystallization and gas exsolution (Sisson and Bacon, 1999). From the bubble sizes of 100–200 μm diameter driving melt through 20–80-μm-wide channels in the crystal mush (Fig. 1H), and a mean gas expansion (∼melt expulsion) rate of 0.07 μm3/s of our experiments, we find ∇P = 0.1–1 MPa/m (after Anderson et al., 1984). Therefore, in a H2O-rich (3 wt%) and SiO2-rich system at 800 °C and shallow pressure (<1.5 GPa), ηmelt ∼103 Pa·s and gas-driven filter pressing could expel melt at v of 0.03–0.3 m/yr (μ = 0.25) and 0.05–0.5 m/yr (μ = 0.4).

In natural silicic systems, the mean size of the dominant phenocrysts (feldspar, hornblende, biotite, quartz) is ∼3 mm (Bachmann and Bergantz, 2004); so, for ∇P = 0.1–1 MPa/m, expulsion velocities of 0.6–6 m/yr (μ = 0.25) and 1.1–11 m/year (μ = 0.4) are expected. This implies that the segregation of crystal-poor melt bodies of tens to hundreds of meters thick can occur within a century, and could act as an efficient method of enhancing segregation compared to the longevity of crystal mushes (104–105 yr; Bachmann and Bergantz, 2004), provided crystallization and volatile exsolution occur sufficiently rapidly to maintain ∇P without reaching the close packing of phases (∼74 vol%), which impedes melt segregation. However, there is little geophysical evidence for the presence of large crystal-poor bodies in the present-day crust (102–103 km3; Bachmann and Bergantz, 2004). Thus, the segregated volumes of crystal-poor caps we infer would generally be below the detection of high-resolution local seismic tomography (cell volume of 125 km3; e.g., Miller and Smith, 1999). And, if extracted, these small volumes of silicic melts may become highly hazardous due to their large volatile content (i.e., H2O dissolved in the melt and exsolved gas) and low ϕ.

Despite the complex interaction between gas exsolution, crystallization, and viscosity, the results of this study serve as a general indication of the effectiveness of gas-driven filter pressing in nature. Our in-situ experiments show that gas-driven filter pressing can operate efficiently in shallow felsic crystal mushes (crystal volume fractions of 0.6–0.7 and below the maximum packing fraction of bubbles and crystals, with melts bearing ∼3 wt% dissolved H2O), and is therefore a viable, though limited, mechanism to rapidly extract large volumes (>102 km3) of hazardous gas-rich crystal-poor magmas within Earth’s crust.

The EU Transnational Access Programme (CALIPSO; number 312284; FP7/2007-2013), the European Research Council Advanced Grant CRITMAG, and EXTREMA European Cooperation in Science and Technology (COST) Action MP1207 supported the work. We acknowledge C. Clapham, D. Hawley, U. Graber, G. Mikuljan, and G. Robert for technical support; D. Baker, M. Polacci, L. Caricchi, and D. Giordano for discussions; and R. Holdsworth, D. Floess, and two anonymous reviewers for their helpful comments.

1GSA Data Repository item 2015243, Tables DR1 and DR2, Figure DR1, and Movies DR1 and DR2, is available online at www.geosociety.org/pubs/ft2015.htm, or on request from editing@geosociety.org or Documents Secretary, GSA, P.O. Box 9140, Boulder, CO 80301, USA.