Deformation T-Cup: A new multi-anvil apparatus for controlled strain-rate deformation experiments at pressures above 18 GPa

Quantitative material property measurements at high pressures and temperatures are of interest in physics and chemistry as well as in the Earth, planetary, and materials sciences. Determination of the mineralogy deep within the Earth, as well as understanding of mineral compressibilites, the effects of pressure, temperature, and chemistry on phase transitions, and so forth have all come from using hydrostatic pressure as a tool to create extreme conditions. Pressure is an important state variable because more extreme changes in physical properties can be achieved using pressure than temperature: for many materials thermal expansion to the melting temperature does not exceed 10%, while compressions of 50% or more are possible. Dynamic properties such as rheology cannot, however, be measured under static conditions prompting recent efforts to develop techniques which are capable of making these measurements under extreme conditions. Apparatus capable of controlled deformation under extreme conditions allows high-pressure phases to be subject to dynamic property measurements under conditions of their thermodynamic stability.

The confinement of samples, in solid media devices, to significant pressures has been possible since the ground breaking work of Bridgeman in the early 20th century.¹ With his apparatus Bridgeman performed experiments at pressures up to about 10 GPa. The subsequent development of the diamond-anvil cell⁴⁵ increased the pressure range over which experiments could be performed. Early attempts to deform samples at high pressure were conducted using the diamond cell as a deformation device;²³ a technique which is still used at very high pressures.³¹ More recently, hard pistons were added to nominally hydrostatic multi-anvil assemblies^3,17 to exert a differential stress onto a sample and deform it to shear strains ∼1. Both these approaches have the limitations of: (1) low total strain achievable, (2) increases in strain are convolved with increasing pressure, and (3) inability to control the timing of strain in the pressure-temperature-time path of the experiment. However, suitably designed static devices can be used with high energy X-rays or neutrons to measure processes active during deformation at high pressure in stress-relaxation experiments.^5,12,13,28 The strain-rate in stress-relaxation experiments is not constant, making the assumption of steady-state that is implicit in quantifying the data difficult to justify and the data more difficult to interpret.

A significant step in overcoming these limitations came with the development of the Deformation-DIA.^7,44 The key innovation in the D-DIA is the addition of secondary actuators within the DIA tooling which enables two of the anvils compressing the cubic sample assemblage to be advanced independently of the other four. This allows deformation to be decoupled from pressure generation. In the decade since its development the original D-DIA design has become a work-horse apparatus for controlled strain-rate deformation experiments, up to pressures around 10 GPa (around 330 km depth in the Earth) and temperatures regularly above 1500 °C. Larger versions of the D-DIA (e.g., MADONNA-1500 in Japan) or cubic presses with 6 independent actuators,³⁰ using the “MA6-6” anvil assembly,³⁴ have been used to successfully increase the pressure range of cubic presses to 25 GPa in hydrostatic experiments²⁰ (∼720 km depth in the Earth) and ∼20 GPa (∼575 km) in deformation experiments.²² The D-DIA was explicitly designed to be used with synchrotron X-radiography and X-ray diffraction, which enables direct measurement of the bulk and elastic strains the sample experiences during the experiment.^{12,15,27,28,37} From these data, the strain-rate and stresses can be calculated which are used subsequently to determine flow-laws, anelasticity, and other dynamic properties of the sample. The D-DIA is not restricted solely to use with synchrotron X-rays. For example, Walte et al.⁴³ looked at the distribution of melts in deformed poly-crystalline olivine in recovered samples and Howell et al.¹⁰ investigated the textures developed during plastic deformation of diamond.

High-pressure controlled-deformation experiments have also been reported in opposed-anvil rotational devices.^2,46 In this style of device one of the anvils is rotated relative to an opposed stationary anvil, the sample is sheared between them and large strains imparted to the sample. However, they suffer from the significant drawback that their operation and interpretation of the results are more complex than those of the D-DIA and 6-axis devices. Nevertheless, results from deformation experiments in the Rotational-Drickamer Apparatus (RDA) at pressures up to 20 GPa have been reported.^16,21

Greater hydrostatic pressures can be generated using Kawai-type apparatus¹⁹ than in DIA type devices. Variations on the original Kawai-type design^35,41,42 have made this style of apparatus more user-friendly and the most common multi-anvil apparatus in use today. With their octahedral sample assembly, these devices are capable of generating pressures up to around 25 GPa with Tungsten Carbide anvils and to pressures in significantly above 35 GPa using sintered diamond anvils.^18,40 Despite this success, the Kawai-type apparatus has not previously been modified into a device capable of controlled strain-rate deformation experiments. Here, we report a new variation of the Kawai-type geometry, the Deformation T-Cup, which has been modified to facilitate controlled strain-rate deformation experiments.

The Deformation T-Cup (DT-Cup) apparatus is a development of the T-Cup design^25,26,41 (Figure 1). The T-Cup itself is a miniaturised Kawai-style apparatus, specifically designed for use with synchrotron X-rays and without any active deformation capability. It has a first stage of six hardened steel wedges, cut from a cylindrical volume. When placed together these first stage anvils form a central cubic cavity (edge length 19.5 mm) with the [111] direction of the cavity aligned along the cylinder axis, which is also the compression axis of the press (Figure 1). This first stage compresses eight 10 mm second-stage cubic anvils. These second stage anvils each have a truncation on their inner-most corner, defining an octahedral volume which contains the sample assembly. This sample assembly is larger than the octahedral void created by the anvil truncations thus the force applied to the first stage wedges is transferred to the assembly and high pressures are generated. Pyrophyllite gasketing is used to contain the sample assembly and improve cell stability at high pressure, although this comes at a cost of reduced pressure generation. The cubic inner anvils are generally made from X-ray opaque tungsten carbide (WC) anvils. Alternatively, X-ray transparent cubic boron nitride (CBN), sintered diamond,²⁴ or, more recently, slotted tungsten carbide anvils⁴ are used, in conjunction with synchrotron X-rays, to enable radiographic imaging and the capture of full powder diffraction rings. The parts of the gasketing and pressure medium in the beam path, up-stream and down-stream of the sample, can be replaced with boron epoxy windows in order to remove unwanted diffraction peaks arising from the gaskets and pressure medium.³³ 

In the DT-Cup, the deformation capabilities come from redesigning the anvils oriented along the compression axis of the press (“top” and “bottom” anvils in Figure 1). The new axial anvils are cubes which have been extended along the [111] axis away from the truncated, inner, corner to make “hexagonal anvils” (Figure 2). The first-stage wedges are modified to create a hexagonal hole to accommodate the modified anvils (Figure 3(a)) while the confining rings are unmodified from the T-Cup. In the absence of deformation, the hexagonal anvils occupy the same volume in the wedges as the cubic anvils they replace and the modified design behaves identically to the original T-Cup design. The redesigned tooling is hosted in a modified 4.06 MN, V7, Paris-Edinburgh load frame (Figure 3(b); for a general description of the V7 load frame, see, Le Godec et al.^25,26). The 150 mm diameter main piston and the breech of the press are altered to hold the 50 mm diameter secondary actuators which back the hexagonal anvils and are used for deformation. These secondary actuators are limited to a maximum of 3.1 kbar by the external plumbing and are able to exert a maximum force of 0.61 MN. The main piston and the two differential pistons are connected to three separate hydraulic pumps, which are independently controlled (Figure 3(b)).

Samples are heated by passing an electrical current between the wedges via the hexagonal anvils and assembly which contains a resistance furnace (e.g., Figure 4(b)). The hole in which the hexagonal anvils sit is a clearance hole and at high loads the wedges dilate sufficiently to break any electrical connection between the wedges and the hexagonal anvil. Heating of the assembly is therefore enabled by shorting the outside ends of the hexagonal pistons to spacer plates which the wedges sit on with copper braid. Thin sheets of plastic between the spacer plates and press as well as behind the hexagonal anvils insulate both sets of wedges and the hexagonal anvils from the press frame.

Two DT-Cup presses have been built, with minor differences between them. One has been deployed on X17B2ss, the monochromatic side-station beam-line to X17B2 at the NSLS, Brookhaven National Laboratory, NY and the other is at University College London (UCL). The DT-Cup at X17B2ss uses the same 10 mm cubic anvils used in the original T-Cup and hexagonal anvils that are 14.13 mm (⁠

$10×2$⁠) across the face-to-face diagonal of the hexagonal shank. The cubic void in the first stage wedges has a 19 mm edge length. No displacement transducers have been fitted to the apparatus because X-radiography is used to monitor the displacement of the differential pistons. The UCL DT-Cup uses larger 16 mm cubic anvils, hexagonal anvils that are 22.71 mm across the face-to-face diagonal and displacement transducers have been fitted to monitor displacement of the differential actuators. The larger anvils require a larger void in the first stage wedges, which in the UCL DT-Cup has an edge length of 29.5 mm.

The changes to the Kawai geometry, discussed here, are analogous to the modifications made to the DIA by Durham et al.⁷ and Wang et al.⁴⁴ in building the D-DIA. As with the T-Cup and the D-DIA, the DT-Cup is designed to be used in conjunction with synchrotron X-rays thus enabling direct monitoring of the strain and stress in the sample during an experiment.

Both DT-Cups have been successfully calibrated and tested. Synchrotron based deformation experiments have been performed at pressure up to 10 GPa, at the NSLS’ monochromatic beam-line X17B2ss and offline deformation experiments have been performed at the pressures of the insulator–metal phase changes in ZnS (15.5 GPa) and GaAs (18.8 GPa).³⁶ A summary of the experiments discussed herein is presented in Table I. In the testing to date, the rate of anvil failure is approximately the same as that in the T-Cup or any other multi-anvil using small sized cubes and the cell design presented here is highly robust to blow-outs, with only one experienced so far in over 20 experiments.

In all the experiments, the behaviour of the two independent differential pistons is very similar and the pressures in each differential actuator track each other closely. In the following discussion, therefore, the differential systems are presented as a single system although there are really two systems performing identical actions. The differential oil pressure presented is the mean of the pressures in the top and bottom pistons, the differential oil volume is the sum of the oil volume displaced by the pumps and the oil flow rate is the time derivative of this. The displacement of the differential anvils is the combined displacement of both the top and bottom anvils.

The sample assemblies used in the DT-Cup commissioning experiments (Figure 4) are similar to those used in traditional Kawai-type apparatus. Each assembly consisted of a 7 mm edge-length, MgO, octahedral pressure medium which was compressed by anvils with 3 mm truncations; generally referred to as a 7/3 assembly. In the ZnS and GaAs experiments the samples were mounted directly in the MgO pressure medium and electrically connected to the axial anvils by 2  mm long, 1 mm diameter steel rods (Figure 4(a)). For the X17B2ss experiments, a 4.7 mm long, straight-cylindrical graphite furnace with a 1.2 mm internal diameter and 200 μm wall thickness was surrounded by the pressure medium; 0.5 mm thick steel disks on the ends of the furnace provided electrical contact with the anvils and prevented the graphite from coming into direct contact with the hexagonal anvils (Figure 4(b)). Some experiments had additional, 1 mm wall thickness, sleeves of crushable alumina between the furnace and the MgO pressure medium; this substantially improved the pressure and temperature efficiency (see Sec. III C). Each experiment had a sample composed of two cylinders, approximately 0.8 mm long, of compressed powders: one of San Carlos olivine and the other of silica. In the present experiments the silica sample was loaded as amorphous SiO₂, except for the lowest pressure experiment in which a quartz sample was used instead. These two samples were sandwiched between two 0.6 mm long corundum pistons and each of these 4 pieces was separated from each other by 25 μm Platinum foil disks. A thermocouple was inserted through the side of the octahedron to butt up against the outside of the furnace but not making electrical contact. The assembly is oriented in the press with the furnace running between the two hexagonal anvils and aligned with the axis of the press.

For the experiments discussed here, 2 × 2 mm cross-section gaskets with a right-angled V-shaped slot cut in the front to accommodate the corners of the octahedral assembly were used (Figure 4). Boron epoxy windows through the gaskets were used in some experiments to improve the quality of the diffraction patterns and test their effect on the gaskets during deformation. Fibre glass or mica sheets are used between the cubic and the first-stage anvils as is normal in Kawai-style apparatus. In the offline experiments all WC anvils were used but in the online experiments 2 × WC hexagonal anvils, 2 × WC cubic anvils, and 4 × CBN cubic anvils were used; using 4 X-ray transparent CBN anvils is sufficient to enable imaging of the entire sample.¹¹ 

The same basic procedure is adopted for all the experiments discussed here. The assembly is compressed, over a few hours, to the desired end-load with the differential actuators fully retracted. After heating and annealing of the sample, the differential oil pumps are advanced rapidly until the differential pistons are lifted from their retracted positions. The point at which the pistons lift is monitored using X-radiographic imaging at the X17B2ss and displacement transducers at UCL. The sample is then deformed at the desired rate by slowly advancing the differential oil pumps until the desired total strain in the sample is achieved or the desired time has elapsed. Multiple strain-rate or temperature steps can be performed in the same experiment by changing the oil flow rate in the differential pumps or the power applied to the furnace. When the desired total sample strain is reached the differential pumps are stopped, the experiment is quenched and the assembly is decompressed, usually over more than 5 h to preserve the anvils. If recovery of the sample is important the positions of the differential pistons are held constant during decompression by retracting the differential pumps at the rate required to hold the pistons. If the sample does not need to be recovered the differential pumps are retracted to their initial position and then main load removed.

During deformation, the end load on the press is maintained, at a constant value, by bleeding hydraulic oil from the main ram. In response to the differential force and the advance of the differential anvils, the main ram retreats fractionally allowing the cubic side anvils to dilate. The dilation of the cubic anvils is constrained by the geometry of the system.

Throughout the ZnS and GaAs experiments the electrical resistance across the samples was measured. These two semiconductors undergo insulator–metal transitions at 15.5 GPa and 18.8 GPa, respectively,³⁶ during which the electrical resistance drops over the course of a few kilonewton increase in confining force from greater than 10⁴ Ω to less than 1 Ω. The transition pressure is taken as point at which the resistance starts its rapid decrease. In order to asses any changes in pressure during deformation, compression of the assembly was stopped approximately half-way between the “insulator” and “metal” resistance values and held at this value while the samples were deformed. Before the experiment was decompressed, the differential pumps were driven backwards, retracting the pistons, to their initial positions. No heating or annealing steps were performed during these experiments.

For the X17B2ss experiments, following compression the sample was annealed at 800 °C for between 2 and 4 h before deformation. Deformation was carried out at up to three separate differential oil flow rates in each experiment at constant temperature or the temperature was increased step wise at constant differential oil flow rate. During the deformation stage of the experiments X-radiographs were acquired and towards the end of each stage of the experiment X-ray diffraction patterns were acquired from each of the samples and one of the corundum pistons. The radiographic images were collected using a visible light camera and a fluorescent Ytrrium-Aluminium Garnet crystal with an exposure of 5 s at a rate of 1 per minute. Example images from the one of the experiments are shown in Figure 5. The change in position of the marker foils between each image was tracked using a pixel cross-correlation method^12,28 from which the sample strains and strain rates were calculated. The strain and strain-rates referred to in the subsequent discussion are calculated from the combined length of both the San Carlos and silica samples. By taking ratios of the strain-rate in the silica polymorphs to that of olivine the relative strengths of SiO₂-polymorphs can be constrained; we will present these results in a separate paper.

Diffraction patterns were captured on a MAR345 image plate with an exposure time of 300 s. The energy of the monochromatic X-ray beam and the MAR image plate parameters for all the experiments were determined from ceria (CeO₂) diffraction patterns taken before the experiments started and analysed in Fit2D.^8,9 Diffraction patterns from the experiments with boron-epoxy windows in the gaskets were very clean (Figure 6) with diffraction peaks clearly visible and indexable. In the experiments without boron-epoxy gasket windows, however, diffraction from the pyrophyllite gaskets and MgO pressure medium obscured the sample diffraction peaks and it was not possible to reliably calculate pressure or differential stresses from the olivine samples. In all the experiments though unique diffraction peaks from corundum were identifiable and used to determine the pressure from the equation of state of Dubrovinsky et al.⁶ In the 946 kN experiment, which did not use boron-epoxy windows, diffraction peaks from stishovite were also distinct and the pressure was determined from the unit cell volumes using the equation of state of Nishihara et al.³² 

The temperatures used in calculating the pressures were those measured by the thermocouple even though the temperature in the corundum or sample is unlikely to be the same as that measured by the thermocouple. The external thermocouple measures significantly lower temperatures than those experienced by sample¹⁴ and at high temperatures there are significant thermal gradients along the furnace axis.³⁸ The temperature difference between the diffraction volumes and the thermocouple has not been calibrated and therefore the temperature used in calculating the pressure is that of the thermocouple. If it possible therefore that calculated pressures are an overestimate of the actual pressure but this can only be a small amount or the stable phase of SiO₂ would be different.

The pressure generating capabilities of the DT-cup were tested using both the semiconductor phase transitions and X-ray diffraction from the samples and are plotted in Figure 7. The assembly used here is very efficient at generating high pressure at ambient temperatures. Our assembly is significantly more effective than the 7/2 assembly (a 7 mm octahedron compressed by anvils with 2 mm corner) used previously by Liebermann and Wang²⁹ (◯, Figure 7). Le Godec et al.²⁵ used T-Cup tooling in a V7 Paris-Edinburgh load frame to calibrate their 7/2 assembly (✩, Figure 7), which is very close in pressure generating capabilities to the 7/3 used in this study. Theoretically, a 7/2 assembly should be more pressure efficient than a 7/3 assembly for the same load³⁹ but the volume of gasketing present has a significant effect on the pressure generation. Here, we have used much less gasketing than was used in the studies mentioned above. The pressure calibration for the DT-Cup using standard 8/3 assemblies and gaskets is very similar to the calibration of the same assembly in a Walker-type multi-anvil.

After annealing at 800 °C, diffraction patterns from the SiO₂ samples in the experiments at 113, 390, 445 and 946 kN (SiO_40, 38, 43, 40, respectively, Table I) show that they were quartz, coesite, coesite, and stishovite. The quartz and coesite patterns were not clean enough for the pressure to be calculated from them but it was possible to calculate the pressure from the stishovite sample which gave the same pressure as the corundum. Another experiment at 951 kN (Fe_15) synthesised hcp-Fe at 500 °C; no bcc or fcc-Fe peaks were observed in the diffraction patterns. The pressure in this experiment was calculated to be 12.5 GPa. In experiments SiO_38 and 39, there are significant reductions in pressure after annealing the samples at 800°C (Figure 7, △). The decrease in pressure is due to a large furnace in an assembly predominantly made from poorly-insulating MgO and the densification of the large amorphous silica sample as it transforms to stishovite or coesite. The pressure loss is significantly lower in SiO_40 because the starting sample was quartz rather than amorphous SiO₂ and in the experiments where crushable alumina insulation was added to the cell (▼, ▲ Figure 7).

The deformation experiments with ZnS and GaAs samples (DT14-004 and -002, respectively) used the electrical resistance of the semiconductors as a probe of the pressure during deformation. The same systematics were observed during deformation of both samples; here the GaAs experiments are discussed and Figure 8 shows the resistance, differential forces, and oil volumes during this experiment. Compression of the assembly was stopped approximately half way between the “insulator” and “metal” resistance values at 1436 kN, when the pressure in the sample was a little above the 18.8 GPa transition pressure. At this confining force the differential pumps were advanced moving the differential actuators. During the initial rapid advancement of the differential pumps there is little movement of the differential actuators but after ∼12.5 cm³ of oil has been displaced the differential pistons advance more rapidly. This is the point at which the differential actuators are floated and the large displacement of the differential anvils is accompanied by a reduction of the resistance across the sample. After a total of 0.47 mm of shortening had been imparted to the assembly (∼15 % strain in the compressed octahedron) the differential pumps were retracted during which the resistance across the sample remained constant. The sample resistance only recovers to its “insulator” value during decompression at a confining force of approximately 300 kN. The decrease in resistance during decompression of the assembly demonstrates that the sample pressure did not drop significantly during deformation and the pressure during deformation was therefore above 18.8 GPa.

In the X17B2ss experiments, the displacement of the differential actuators is not measured directly but strain in the sample, calculated from the X-radiographs, is used as a proxy for displacement of the differential anvils. The strain from X-radiographs acquired immediately before and after the diffraction patterns was averaged to give the mean strain in the sample during the diffraction acquisition. The change in pressure as a function of strain can then be determined, by plotting pressure against strain (Figure 9). At constant temperature, the pressure during deformation is constant to within the spread of the data. In SiO_39 (946 kN) the calculated pressures in the stishovite and corundum are within error of each other and roughly constant at 9.2 GPa and in SiO_40 and _38 (113 and 390 kN) there is little pressure drop even at strains in excess of 50%. However, in the experiment where the temperature was increased stepwise the pressure decreases with increasing strain and temperature because there is a general thermal relaxation of the pressure medium and gaskets. There is no indication in the data of a strain-rate dependence to the pressure.

The forces on the pistons and the temperature profile for a typical experiment are shown in Figure 10. While the deformation rams are advancing, the force exerted by the main piston is kept constant but the force exerted by the differential rams (and the differential oil pressure) increases significantly. The differential actuators are housed within the main-ram and breech of the press and act both to strain the sample and against the confining force provided by the main-ram. Balance of forces requires that the force exerted by the main ram on the non-differential cubic anvils must therefore decrease, enabling them to dilate out of the way of the differential anvils. The increase in force on the differential anvils and corresponding reduction on the cubic anvils balance out to keep the sample pressure approximately constant as a function of strain (Figure 9). In the current experiments movement of the wedges was not measured but gap between the cubic anvils can be measured from the radiography images. In experiment SiO_38 (Figure 5), the anvil gap has increased by approximately 2% at 20% strain and by ∼14% at 55% strain.

We note here that the differential force is not a function of strain-rate but is a reproducible function of strain. This suggests that the cause is internal friction in the cell and not pressure-lag in the oil lines. This is further confirmed by the fact that when the differential pumps are stopped at high forces there are only trivial changes in the position of the pistons. If the system was over-driven it should be expected that the differential pistons continue to advance after the pumps have stopped. The behaviour of the differential system is consistent with that observed in the D-DIA, although the rate of increase in the differential force with strain appears to be much greater in the DT-Cup as would be expected since 6–8 geometries have significantly more gasketing than DIA geometries.

The total strain in the sample, used here as a proxy for the advance of the differential pistons, is a function of oil volume displaced by the differential pumps and the end load (Figure 11(a)). Higher end loads require a greater volume of oil to be displaced before the differential anvils first move and strain in the sample can be observed X-radiographically. The amount of strain per unit volume of oil decreases with increasing end load. At constant temperature, the strain-rate in the sample is an approximately linear function of oil displacement rate (Figure 11(b)) and, for a constant oil displacement rate, decreases with increasing end load. At constant differential flow rates higher temperatures correspond to faster sample strain-rates as more of the assembly-strain partitions into the hot-part of the assembly.

The volume of oil needed to initially displace the differential pistons scales with the end load (Figure 12). For the standard amount of gasketing used in the experiments here the force required to move the differential pistons is 0.30 of the confining force. The forces required to lift the differential actuators in the GaAs and ZnS experiments was ∼0.24 times the confining force. The lower value for floating the differential actuators in the semi-conductor experiments is because in these experiments the first motion of the differential actuators can be observed directly using the displacement transducers (as in Figure 8) but in the X17B2ss experiments the actuators have to have moved a significant distance before the displacement can be observed, by eye, in the X-radiographs. Theoretically, the force required to advance the differential actuators in the octahedral geometry is a quarter of the confining force. This is consistent with the experimental values measured here and with D-DIA where the force required to move the differential pistons is approximately 1/3 of the confining-force.

The results presented here demonstrate that the DT-Cup variation on the Kawai-type geometry can be used for deformation experiments in a manner similar to that which the deformation-DIA geometry has been for the past decade. It is a viable apparatus capable of performing deformation experiments at pressures above 18 GPa at modest end-loads and is reliable and reproducible in its experimental operation. The rate of experimental and anvil failures are not significantly greater than that in the T-Cup or other static Kawai devices, although the number of experiments currently performed in the DT-Cup is small.

The DT-Cup devices discussed here are limited in the volume of sample they can act on by the small size of the anvils. The pressure at which samples can be deformed is limited by the maximum force available in the differential actuators which in turn is determined by their small diameters and the maximum oil pressure of the control system. However, future developments of larger DT-Cup style machines will overcome many of the limitations of the current generation of devices.

S.A.H. was funded by NERC post-doctoral research fellowship (NE/H016309/1). S.A.H. and L.L. were funded by NSF Grant No. EAR0809397. R.J.M. was funded by Marie Curie Research Training Network “c2c” Contract No. MRTN-CT-2006-035957. Neil Hughes is thanked for his technical support for the project and Figure 2. Use of the National Synchrotron Light Source, Brookhaven National Laboratory, was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-98CH10886. Use of the X17B2 beamline was supported by COMPRES, the Consortium for Materials Properties Research in Earth Sciences under NSF Cooperative Agreement EAR 10-43050 and by the Mineral Physics Institute, Stony Brook University.

We thank the two anonymous reviewers for their comments which resulted in a much improved paper.