The samples with abundant gaseous fluid inclusions had very high strengths and did not reach a steady state flow stress, indicating that recovery was inhibited in these samples (Fig. 1, Selverstone, 2005). Samples with aqueous inclusions had a lower strength under similar conditions. This difference suggests that samples containing gas-rich inclusions behave similar to dry samples, i.e., with a lower recovery rate. This is consistent with the experiments of Post et al. (1996). They showed that dislocation climb and recrystallization is dependent on water fugacity. However, they only showed this dependence in pure annealing experiments and did not directly relate the nature of the fluid to strength. Our experiments do show this direct relationship. Though we cannot quantify the water-fugacity in our samples, this paper is presenting a detailed study on the effect of different kinds of fluid composition on the deformation behaviour of natural quartz with different starting fluid inclusion populations, and allows interpretation of the processes that are affected.

Dislocation creep involves several processes. Dislocations can be generated at micro-scale fluid inclusions (impurity) (McLaren et al., 1989; Vityk et al., 2000) and in natural samples these sources are usually abundant. At conclusion of the experiments all samples show undulose extinction indicating the presence of dislocations (Fig. 2). This suggests that the fluid composition has minor effect on the formation of dislocations.

The second process that occurs during dislocation creep is gliding of dislocations. Our experiments do not show whether fluid composition affects this process. Since deformation is accompanied by recovery processes with rate controlling grain boundary bulging and dislocation climb (compare Hirth and Tullis, 1992) we cannot distinguish an effect of fluid composition on dislocation glide.

The nature of the fluid likely has an effect on both, the dislocation climb rate and the grain boundary migration rate. Since the samples are single crystals there are initially no grain boundaries. Thus, at low strains grain boundary migration cannot contribute to the low strength, so climb is interpreted to be involved. Grain boundary migration can only operate after subgrains are rotated enough to form high angle grain boundaries, or after nucleation of new grains along grain boundaries (Hippert and Egydio-Silva, 1996), or after early stage micro-cracking followed by healing and the formation of nuclei (Tarantola et al., 2000). The samples with gaseous inclusions are stronger than those with aqueous inclusions right from the start. So the fluid composition is interpreted to affect the climb rates of dislocations at this stage, consistent with the data from Post et al. (1996).

Recrystallized grains

The difference in size and number of recrystallized grains between samples containing gaseous and those containing aqueous inclusions (Fig. 2) indicates that grain boundary migration is enhanced by the aqueous phase relative to the gaseous phase. Since the gas-rich samples had a much higher strength it can be assumed that they built up a much higher dislocation density than samples with aqueous fluid inclusions. Higher differences in dislocation densities should enhance the grain boundary mobility, however grain boundary mobility appears much lower or retarded in the presence of the gas inclusions. The fluid is thought to affect the grain boundary mobility in two different ways, related to the assumption that the gaseous phase is CO2-rich and the aqueous phase is water-rich. First the quartz is less soluble in the CO2, thus the flux of quartz across the grain boundary will be lower and the mobility of the grain boundary arguably also lower in the CO2-rich case. The second effect is due to "wetting" of the grain boundaries. H2O -rich fluid will form films along grain boundaries which enhance the grain boundary mobility (Urai, 1986; Rutter and Brodie, 2004). CO2-rich fluids do not form such films but remain in isolated inclusions and may even inhibit migration by acting as a pinning second phase (Herwegh and Berger, 2004). However, as there is no comparison to dry samples provided in this study, the effect of H2O on the flow strength cannot be estimated. The study by Stipp et al. (2006) suggests a negligible effect of H2O on the recrystallized grain size of quartz and, by inference, on grain boundary mobility.

An H2O-CO2 effect was suggested by e.g., Drury and Urai (1990) who proposed that a H2O-rich phase wets the grain boundary during migration, whereas a CO2 -rich fluid does not. We have not observed the boundaries during migration, but some indication of the fluid geometry is given by the SEM observations on the broken surfaces of the samples (e.g., experiment 33, Fig. 3). Samples containing aqueous fluid inclusions are characterized by smooth grain boundary surfaces, whereas the grain boundaries in gas-rich samples are more irregular. This difference in geometry suggests an aqueous fluid film but isolated gas inclusions. Samples with aqueous fluid inclusions have open triple junctions, whereas the gas-rich samples do not. We also see this in areas where grain boundary migration stopped and the boundaries show evidence for static recrystallization (i.e., straight grain boundaries, 120° junctions). Here the fluid has pulled back into isolated inclusions. This difference in fluid distribution dependent on the composition in the statically recrystallized areas of the sample is consistent with the observations of Holness (1993). She showed that at experimental conditions similar to ours the wetting angle of H2O in quartz was < 60°, but that CO2-fluid has a wetting angle of > 60°.

A striking observation is that the recrystallized grains were elongated in the direction perpendicular to σ1. Mostly the elongation of the grains was stronger than that of the total strain ellipse, which indicates that the grains also grew after the bulk shortening perpendicular to the maximum applied stress. There are several factors that may explain this effect. A likely cause is an effect of orientation dependent grain boundary mobility, due to fluid film thickness. As hypothesized by Urai et al. (1986), the migration rate of a grain boundary is dependent on the thickness of the fluid film, which is affected by the film orientation relative to the principal stresses. Drury and Urai (1990) suggested that the fluid film thickness is a function of boundary orientation with respect to stress direction, with thicker films on boundaries parallel to σ1. If our recrystallizing grains showed increasing migration rate with film thickness, a faster migration rate would be expected perpendicular to σ1. For very thin films, the migration rate will increase with film thickness till a certain limit, after which increasing film thickness will result in a decreasing migration rate, due to an increase in diffusion path length. Similar results were found by Karato and Masuda (1989) for grain growth in deformation experiments at high stresses.

A second possible cause for the elongated grains is grain growth similar to an anti-crack mechanism. The small, recrystallized grains are dislocation free and thus weaker than the strain-hardened surroundings. During stress relaxation they will deform more easily and cause stress concentrations in the surrounding material at those edges which are parallel to σ1. The strain energy gradients in those regions then increases, leading to an increase in the driving force for migration and thus faster migration in the direction perpendicular to σ1. However, since the elongation of recrystallized grains is observed in all samples it is unlikely that growth anisotropy is solely responsible for the elongation. In the gas-rich samples the stress remaining after deformation is very high, and thus a slight increase in dislocation density is unlikely to cause very large differences in driving force.

Changes in fluid inclusions

During the experiment the fluid inclusions increase in number and decrease in size, suggesting that they break up during the experiment. We observe this modification in all samples, even the one that has been annealed for a very short time. We also see this phenomenon in most areas of the sample, whether recrystallization occurred or not. These observations suggest that the breaking up of fluid inclusions is not related to the recrystallization, but happens during loading and deformation of the sample by intracrystalline leakage along mobile dislocations (Kerrich, 1976, Hollister, 1990; Vityk et al., 2000; Tarantola et al., 2010; Schmatz and Urai, 2011). Fluid inclusions may develop an internal overpressure and form microcracks (Roedder, 1984; Tullis and Tullis, 1985) that subsequently heal into trails of smaller inclusions (e.g., Fitz Gerald et al., 1991).

Recrystallization causes some important changes in the fluid volume as the recrystallized grains contain very little fluid inclusions, whereas the starting material contained approximately 4 vol.% fluid. This raises the question where the fluid has gone. Even if the recrystallized material contains no optically visible fluid inclusions, some fluid may have been incorporated into the crystal structure as point defects (max. ∼40ppm, Paterson, 1986; 1989), or point defect clusters with sub-microscopic scale fluid inclusions similar to the type that have been observed in milky quartz or amethyst respectively (Kekulawala et al., 1978; Aines and Rossman, 1984). Milky quartz may contain up to 2000 H/106 Si; amethyst contains up to 3000 H/106 Si, which corresponds to ∼1 vol% water. There thus remains ∼3 vol% water unaccounted for that must have been swept by the boundary. This fluid must have interacted with the grain boundary in some way, since the inclusions clearly did not remain in the same position after the boundary has swept by. Thus there must be a lateral fluid pathway on the grain boundary which allows transport of fluid which has been proposed in many studies dealing with the water fugacity of quartz (e.g., Stipp et al., 2006). We may see some evidence for this in the SEM, on the surfaces of samples broken along the grain boundaries (Fig. 3). The open boundaries may heal when the grain boundary stops moving. The fluid will then re-distribute into distinct pores. We don't know the microstructures this process would produce, but features as shown in Fig. 3f could indicate a healing grain boundary.