The standard rim-forming mineral reaction olivine + quartz = orthopyroxene (Ol + Qtz = Opx) has been experimentally performed at very dry conditions, i.e., only 20 wt ppm of water present in the sample container as a fluid, and most “water” was hydrogen dissolved within the solid crystals. Reaction rates and resulting fabrics at Ol-Qtz interfaces mimic reaction features previously known from water-rich conditions (tens of thousands ppm). Our experiment indicates that very small amounts of water (tens of ppm of the entire sample) are highly effective in facilitating mineral reactions where water acts as a catalyst and creates porosity that might start to migrate from the initial centers of self-propagating mineral reaction, if rock deformation allows. The threshold between dry and wet in granulite and eclogite facies rocks is crossed once the nominally anhydrous minerals are saturated with hydrogen.


The effect of water on reaction kinetics in solid-state systems such as rocks has been a conundrum for as long as the principle of thermodynamic equilibrium has been applied to rocks (Goldschmidt, 1912; Eskola, 1920) and as this principle has been extended (minimized) to the principle of local equilibrium (Korzhinskii, 1959; Thompson, 1959). Experimental petrologists have known for many decades that some water is always necessary to allow mineral reactions to occur, even if H2O does not appear in the reaction equations. It has long been emphasized that dry rocks under eclogite facies pressure-temperature (P-T) conditions may remain out of equilibrium for geological times, and that metamorphic reaction pulses might only occur due to episodic and local infiltration of water (e.g., Rubie, 1986; Austrheim, 1987). However, quantification is missing. Experimentalists are especially interested in finding kinetic threshold values between dry and wet systems. Those who study rock deformation have known about hydrolytic weakening for four decades, and there is no doubt that water amounts dissolved in a silicate mineral even in tens of parts per million (ppm) ranges may have a strong effect on plasticity in minerals such as quartz (e.g., Kronenberg and Tullis, 1984) or olivine (e.g., Mackwell et al., 1985). In contrast, for the kinetics of mineral reactions occurring at interfaces, for example between olivine-quartz pairs, the effect of water and the kinetic limits have long remained unconstrained. Notionally dry experiments (e.g., Milke et al., 2001, 2009a) give results for diffusivities in line with wet ones unless special precautions are taken (e.g., the dry thin-film experiments of Milke et al., 2007). Gardés et al. (2011, 2012) performed piston-cylinder experiments at 1.5 GPa, both dry and with a range of added water contents. In the wet experiments, even the run with the lowest H2O content implied diffusivities 4–5 orders of magnitude higher than the dry case, implying a threshold for water-saturated behavior at 500–1000 wt ppm (ppm by weight). This was a first experimentally well-founded value for the water amounts necessary to allow for efficient reaction and growth of common rock-forming minerals at mineral interfaces in solid-state systems similar to metamorphic rocks.

Here we provide a three-dimensional (3-D) image of the quartz-orthopyroxene (Qtz-Opx) interface from a rim growth experiment between single crystals of olivine (Ol) and Qtz (Milke et al., 2009a), and especially visualizing the pores at this interface, and their quantitative data yield. Our observations, made by the novel technique of focused-ion-beam 3-D imaging, allow new insight into the mechanisms of mineral growth at interfaces. They are complementary to the observations of Gardés et al. (2012), and demand further consideration of the amount of water necessary to make a dry silicate system wet. We provide evidence for a mechanism of mineral growth that creates and uses microporosity or nanoporosity under fully hydrous conditions despite very low water concentration in the total sample or rock.


Our Opx rim growth experiments have been made using alternating stacks of oriented Qtz and Ol single crystal discs enclosed in gold capsules in a piston-cylinder apparatus (for details, see Milke et al., 2009a). No water has been added to the cold-sealed Au capsules, but water entered them uncontrolledly during the course of the experiments (T gradient from 950 to 700 °C along the 8-mm-long capsules; P = 1.9 GPa), and was most probably present early (for an analysis of water access to such initially dry experiments, see Joachim et al., 2012). The 3-D tomographic image was made using the slice-and-view technique in an FEI Quanta 3D FEG DualBeam, where a Ga ion beam was used for cutting and an electron beam was used for imaging. The sample was cut from the polished specimen KK7 from Milke et al. (2009a), and displays a reaction product between natural Qtz and Ol. It belongs to the parts of the rim that were named “bulging” (see Milke et al. 2009a) and were related to the local presence of an aqueous fluid. This specimen contained 5 single crystal discs, each 1 or 2 mm thick, that made up 4 initial Ol-Qtz interfaces. The viewed Opx-Qtz interface was ion-beam thinned by 100 nm 50 times, and photographed at each step using the secondary electron signal. After completed data storage, the images were combined into a tomographic image using ImageJ-Fiji software. Minor corrections to the three gray-scale classes Qtz, Opx, and pore were done by hand if necessary.


The Opx-Qtz interface is made up of elongate Opx crystals piercing the Qtz single crystal substrate. This growth feature has been observed in experiments with uncontrolled water content (e.g., Yund, 1997) and was replicated by Gardés et al. (2012, their figures 4A and 4B) in experiments loaded with 1 or 5 wt% water. Figure 1 shows Opx-Qtz interfaces from experiment KK7, including the exposed interface block for 3-D imaging (Fig. 1A), its surface after four times slicing 100 nm away (Fig. 1B), two representative views of water-rich sections of the Opx rims at 853 ± 10 °C (Fig. 1C) and at 901 ± 10 °C (Fig. 1D), and a sketch of the sample setup (Fig. 1E). The dark areas labeled active pores in Figure 1B indicate continuous channel-like porosity from the Opx crystal tip toward its base. The 3-D tomographic image is displayed in Figure 2.

From Figure 2, it becomes clear that (1) fluid-filled porosity is created at the Opx-Qtz interface and provides fast transport pathways; (2) the fluid-filled pores extend continuously as channels from the crystal tips of the propagating reaction rim toward its main body; and (3) the propagation of active pores (i.e., extending continuous fast diffusion pathways) at the ultimate mineral interface is paralleled by leaving behind passive pores (i.e., isolated) in the reaction rim.


Opx crystals paralleled by a pore channel and piercing Qtz clearly indicate the action of water, as an aqueous fluid is needed to keep pores open under the confining pressure of 1.9 GPa. Such pores are absent in really dry experiments at ambient or high pressure (e.g., Milke et al., 2007, their figure 4; Gardés et al. 2011, their figure 3). A lucky cut from our water-poor experiment is displayed in Figure 1B, where the pore channel extends perfectly straight within the sample surface. Any porosity in our single-crystal reactant samples is restricted to the Opx reaction rims. The pores are situated at the Opx-Qtz interface, where the total negative reaction volume manifests when SiO2 is mobile (Milke et al., 2001). From these observations, the growth mechanism of orthopyroxene at the Opx-Qtz interface can be deduced, and we explain why the Opx-Qtz interface is rugged in all the numerous water-bearing experiments, and smooth in all dry experiments (Yund, 1997; Fisler et al., 1997; Milke et al., 2001, 2007, 2009a, 2009b; Abart et al., 2004, Götze et al., 2010, Gardés et al., 2011, 2012). Once a water-bearing pore comes into existence at this interface, it provides a fast diffusion pathway and thereby allows enhanced growth of Opx. Crystals with their fastest growth direction vertical to the initial interface (along c for Opx) have a geometric advantage and after a short time dominate the reaction front. Electron backscatter diffraction analysis has shown that large crystals also form parallel to the Opx-Qtz interface (Abart et al., 2004). All of this can be explained as growth along an interface-parallel fluid-filled pore. Directions other than parallel (the easy way) or vertical (the competitive way) to the reaction front are less favored. We sketch the situation at the Qtz-Opx interface in Figure 3. There is a permanent flux of dissolved SiO2 through the Opx rim to the Opx-Ol interface, where this SiO2 is consumed to produce Opx. In return, MgO (or FeO or other) diffuses along grain boundaries through the Opx rim and along the fluid-filled pore channel to the tip of the orthopyroxene crystal following a chemical potential gradient. It has been demonstrated (Milke et al., 2011) that on a 1 μm scale, local equilibrium with respect to Fe, Mg, and Ni is established at the Opx-Ol interface. It is thus probable that a(SiO2) at the Opx-Ol interface is also buffered by local equilibrium.

At the Qtz-Opx interface the situation is different, because Qtz is almost pure SiO2, and because here the solid reactants interact via the pore fluid. The absence of a buffering equilibrium between Opx and Qtz with respect to the divalent ions leads to uncontrolled Fe enrichment in Opx (Milke et al., 2011). The narrowness of the pores next to the propagating crystal tips prohibits any equilibration with respect to silica, because the distance between the Qtz and Opx interfaces is less than the thickness of the interface boundary layer, and promotes autocatalytic dissolution-growth coupling (Anderson et al., 1998a, 1998b). Thus, a(SiO2) is probably higher than in a bulk fluid in equilibrium with Qtz, and the gradient in a(SiO2) from the crystal tip to its base is probably very shallow. We conclude that a steady-state flux of SiO2 from the porous Qtz-Opx interface toward the Ol-Opx interface through the Opx grain boundary network is established that is balanced by an MgO (FeO) flux in the opposite direction, and that it is pointless to name dissolution, transport, or crystallization as being rate limiting, because all these steps are interrelated.

Experimental observations of replacement fabrics pointed out that reaction-induced porosity is a prerequisite for replacement reactions (Putnis, 2002). Figure 2 reveals that the pores along crystals protruding from the Opx rim into Qtz typically are crystal-parallel channels. Evidently, this specific growth mechanism keeps those channel-like pores open and preserves their permanence as a diffusion shortcut. In the given example the slight oversaturation needed to allow Opx growth at its (001) plane is just reached at the Opx crystal tip, but not at the (100) or (010) planes (thus needle-like growth). Channel-like porosity also evolves in solvent-aided phase transformations, and its experimental study in halogenide (Raufaste et al., 2011) and alkali-feldspar systems (Norberg et al., 2011) has led to the suggestion that porosity formation of this kind might occur widespread in replacement reactions. The reaction studied here can be understood as a replacement reaction (constant volume replacement) at the Ol-Opx interface plus overgrowth in transient porosity at the Qtz-Opx interface (Milke et al., 2001), and therefore provides opportunity for extending the ideas from phase-transformation experiments to more complex metamorphic crystallization.

From the overall negative reaction volume of Ol + Qtz = Opx (∼−6%), and the constant-volume replacement at the Ol-Opx interface, it follows that transient porosity is formed at the Qtz-Opx interface. For the real pore volume, the solubility of the solid phases in an aqueous fluid might be even more important (Putnis, 2002). At elevated pressures and temperatures, SiO2 is soluble in H2O (providing ∼20% of the fluid at the conditions of our experiment; Manning, 1994). It is thus inferred that in the silicate crust and mantle of the Earth, mechanisms of mineral growth as sketched here evolve in many situations, where small amounts of water are attracted to transient porosity and inevitably lead to self-propagating reaction progress.

Our experiments demonstrate that Opx crystallizes in an identical way in upward or downward directions, and that its formation is clearly independent from gravity (Figs. 1C and 1D). Opx formation is only controlled by the chemical interaction (surface reactions and diffusion) between the reactant and product phases via the fluid film. This would certainly be different if concurrent deformation had occurred and the solid minerals had competed with the fluid in the gravity field. In this case, the fluid would have had an upward-moving tendency due to its lower density. In our experiment there was practically no deformation of the single crystal reactant discs as would have been necessary for the compaction-driven expulsion of metamorphic fluids. The observations from our experiment thus reflect the initial conditions of pore formation before they start upward migration in deforming rocks. In the case of pore formation presented here, no water is lost by pore migration, and the Opx-forming reaction occurs as a self-propagating process permanently creating new transient porosity, where water acts just as a catalyst (Rubie, 1986).

From the 3-D mapped volume, the total porosity within the samples and thus the fluid volume can be qualified. The total amount of pores in the 3-D sampled volume is 0.65 vol% (much lower than the solid volume change of the reaction and thus representing just the fluid-filled volume). The sample volume contains four Opx reaction rims at Ol-Qtz interfaces. Their overall similarity on backscattered electron images of the complete reaction interfaces (Milke et al., 2009a) suggests that all of them have a similar average porosity. At 1.9 GPa and 1000 °C, the density of water in equilibrium with Qtz is ∼1.5 g/cm3, and thus half that of the silicates employed, at a water:SiO2 fraction of ∼4:1 (Manning, 1994). Each vol ppm (ppm by volume) of pore space therefore stands for 0.4 wt ppm water present as a free fluid.

By extrapolating the amount of fluid-filled pores from our 3-D image to the entire interface area, its water content would be overestimated, because large parts of the reaction rims (regular rims in Milke et al., 2009a) are devoid of porosity. However, there might be room for fluids between the reaction sample and the walls of the Au capsule. Approximately two-thirds of the total interface area in the samples consists of Au-Qtz or Au-Ol interfaces. Milke et al. (2009a) showed that only very little Opx formed at Au-Ol interfaces, and that in most cases the Opx rims do not extend above the initial Qtz-Ol contact. This is clearly different from rock-cylinder experiments formed at fluid-rich conditions; i.e., where a continuous fluid film surrounded the entire sample. Reaction rims produced under such conditions form at constant thickness on each type of substrate mineral (Metz and Milke, 2012). There is thus very little doubt that, in general, no fluid was present between the inner wall of the Au capsule and the reaction sample, and that practically all water belonging to a free fluid phase inside the sample container was restricted to the reaction rims in the interior of the sample. The total amount of fluid at Au-Qtz and Au-Ol interfaces surrounding the sample therefore was within the same order of magnitude as in the pore-free parts of the Qtz-Ol interfaces, or lower. Extrapolating the amount of water in the sampled volume (that was selected as to represent a fluid-rich area) to the total sample seems to be a fair assumption that overestimates rather than underestimates the total free water content in the sealed capsule, and allows this simple calculation: (thickness of rims × number of rims/length of sample) × percentage of porosity = total pore fraction of the specimen; i.e. (20 μm × 4/8 × 103 μm) × 0.65 vol% = 65 vol ppm. Converting vol ppm H2O to wt ppm H2O by a factor of 0.4 leads to ∼21 wt ppm H2O in the capsule, present as a free fluid. The total amount of water in the samples is the sum of intracrystalline water and water in pores. After an experiment, ∼80 wt ppm water in San Carlos Ol and 20 wt ppm in Qtz were determined (each representing 50 vol% of the total sample; Milke et al., 2009a), that mostly comprised hydrogen defects within the crystal lattices. Both phases were saturated with intracrystalline water. The total amount of water in crystals and pores of the sample KK7 was therefore ∼70 wt ppm, where ∼50 wt ppm were dissolved in the solid crystals (as nonmolecular water) and ∼20 wt ppm existed as molecular water in pores.

Experiments on Opx rim growth between Ol and Qtz single crystals (Milke et al., 2009a) produced growth features identical to those shown by Gardés et al. (2012), although in their powder experiments much more water was present. The Opx growth rates in our experiment are equal to those from powder experiments at water-rich conditions (to 5 wt%; e.g., Yund, 1997). Thus, it is concluded that locally at initial Ol-Qtz interfaces our samples were oversaturated in water and provided aqueous growth conditions, despite their overall minuscule water content. Powder experiments on Ol + Qtz = Opx without intentional addition of water contain ∼2000 ppm H2O due to surface adsorption (Milke et al., 2009b), still much more than in our experiment. This difference can be explained by the much smaller area of initially water-undersaturated grain boundaries in our single-crystal-discs experiment, that diminished the driving force for water diffusion into the sample and localized the action of water. Hence, the total amount of water present in an experimental or natural system is not a meaningful measure for its action on reaction kinetics, as long as it is not related to the existing interface area (in our experiment, 1 wt ppm water as a fluid per 10 mm2 interface area). Our observations indicate that even very small amounts of water might have a strong effect on mineral growth, for example along fractures in otherwise dry granulite or eclogite facies rocks. We thus conclude that once the nominally anhydrous minerals hosting the fracture are saturated with intracrystalline water, only a very few ppm of excess water on grain boundaries have the potential to enhance the growth rates of transport-controlled reactions by several orders of magnitude. It is therefore the grain size (i.e., grain boundary volume) and the ability of nominally anhydrous minerals to take up hydrogen that determine if silicate mineral systems (rocks) behave as dry or wet systems. Once the nominally anhydrous minerals are hydrogen saturated the system turns to wet. This is different from rock deformation, where hydrolytic weakening is highly effective even before hydrogen saturation is reached. Joachim (2011) reported rim growth rates between H-saturated periclase and Qtz in an overall water-undersaturated system. The reaction occurred ∼10 times faster than in completely dry state, but ∼1000 times slower than at water supersaturation (if only by a few ppm). It remains an interesting question, if fine-grained rocks under deformation might have a higher potential to soak up aqueous fluids within their grain boundaries, and keep them from acting as a catalyst for thermodynamically stable reactions, as compared to coarse-grained rocks or dense brittle rocks with a limited amount of fractures; the experimental record suggests that this might be true.

These experimental findings have important implications for the interpretation of the metamorphic history of dry granulite or eclogite facies terranes by introducing a quantitative concept for the presence of water. Grain boundary volume (grain size) and intracrystalline solubility of hydrogen are identified as the leading variables controlling the occurrence of retrograde reactions in dry rocks. Only very few ppm of water present as a pore fluid, and in excess to lattice-bound water, can lead to massive reaction due its action in autocatalytic dissolution crystallization. Persistence of dry metamorphic assemblages outside their stability field thus means that they were devoid of any aqueous fluid even in the low wt ppm per mm2 interface range, probably because their nominally anhydrous minerals were not saturated with hydrogen.

We thank the DFG for funding the projects MI 1205/2-1 (Milke, Neusser, Kolzer) and AB 314/3-1 (Neusser) within the framework of FOR 741.