In-situ study of microstructures induced by the olivine to wadsleyite transformation at conditions of the 410 km depth discontinuity

Samples

Two starting materials were used: an olivine single crystal and sintered polycrystalline olivine samples. The olivine single-crystal was a fragment of natural San Carlos olivine, with dimensions of ~20 × 20 μm and polished manually to obtain a thickness of 10–15 μm. The polycrystalline samples were prepared from a powder of natural San Carlos olivine, ground in a planetary mill and sintered at 500–600 MPa and 1500 K for 40 min in a piston-cylinder apparatus. A piece of the sintered product was then prepared for electron microscopy and characterized in a FEG JEOL JSM-7800F scanning electron microscope (SEM) using the electron backscattered diffraction (EBSD) technique. This characterization shows that it is a pure olivine polycrystal with heterogeneous grain sizes, ranging between less than a micrometer and 50 μm, and with no preferred orientation (Figs. 1a and 1b). Infrared analysis shows <3 ppm of H₂O in the olivine grains of the sintered polycrystal (Fig. 1c). That analysis was done with a Bruker Hyperion 3000 FTIR microscope coupled with a Vertex 70 spectrometer, which was equipped with a simple MCT detector and a “focal plan array” detector. Once sintered, the product was cut into fine slices with a precision diamond wire saw and mechanically polished to a thickness of 10–15 μm. Finally, samples were cut into disks of 20 μm diameter by lasers (Fig. 1d).

Figure 1

Characterization of the polycrystalline starting material for experiments at ESRF and PETRA III. (a) EBSD band contrast image showing the shapes and sizes of the grains in the olivine polycrystal after piston-cylinder sintering. (b) EBSD orientation map of the starting material. (c) FTIR measurements of water content in grains of the sintered polycrystal. (d) Photograph of the diamond-anvil cell pressure chamber (closed cell, top view). The olivine polycrystalline sample is a 20 μm diameter disk loaded with a MgO pressure medium inside a 140 μm diameter hole in a rhenium gasket.

In both cases, samples were coated with a 150–200 nm thick platinum layer on both sides. This step ensures proper absorption of the heating infrared lasers by the sample. It is also critical to maintain stable temperatures as olivine and wadsleyite have different optical absorption properties. No signal from the platinum coating is seen in the diffraction images because of the very small volume fraction of platinum compared to the sample. Nevertheless, temperature stability during laser heating was drastically improved for samples with platinum coating. Earlier attempts with uncoated samples showed jumps in temperatures of several hundreds of degrees during pressure increases or phase transformations and those data sets were systematically discarded. Such large temperature oscillations did not occur when using samples coated with platinum. The platinum coating is then a convenient way of increasing the stability of the heating without decreasing the diffraction signal of the sample.

High-temperature/high-pressure experiments

To reach the high pressure and high temperature of the 410 km depth discontinuity, we use diamond-anvil cells (DAC) coupled with laser heating (Fig. 2). We used diamonds with 300 μm flat tips, rhenium gaskets with holes of 100–145 μm in diameter to serve as sample chambers (Fig. 1d), and membranes to control sample pressure remotely. As pressure media, we used either MgO, KCl, or alumina (Table 1). All three are chemically inert in an olivine-wadsleyite system at the investigated pressure-temperature conditions (Shen and Lazor 1995; Kimura et al. 2017; Zhou et al. 2020). We can rule out undesired reactions in our experiments as no other phase than olivine, wadsleyite and the pressure media are present in the diffraction patterns (Fig. 3). In all cases, the sample and pressure media were loaded under a controlled argon atmosphere into the DAC to prevent the interaction of the hygroscopic pressure transmitting media with air moisture.

Figure 2

Experimental layout for multigrain crystallography (MGC) measurements at high pressure and high temperature at a synchrotron beamline. View from the top. The synchrotron X-rays beam (thick blue line) passes through the DAC (in green) and is transmitted or diffracted by the crystallites in the sample. The diffracted rays (thin blue lines) form the difraction pattern on the detector (in gray). The DAC is rotated in ω and data are collected at diferent ω values. In the reported experiments at both PETRA-III-P02.2 and ESRF-ID27, the laser heating system was rotating with the DAC during ω rotations. At SOLEIL-PSICHÉ, the lasers were fixed and the sample was quenched in temperature before collecting MGC data.

Figure 3

Experimental difraction data for data set P2_01_s07 where olivine and wadsleyite coexist. (a) Difraction spots extracted by the 3D-XRD process from the whole difraction collection, plotted in azimuth vs. inverse d-spacing and staked in omega. (b) Integrated difraction pattern with annotated diffraction peaks for KCl, olivine, and wadsleyite. Here, the difraction pattern is plotted as a histogram of locations of the difraction spots shown in (a) vs. inverse of the d-spacing. Using such a histogram leads to sharp difraction signals and allows for easy sample identification.

The state of the sample was followed using in situ multigrain X-ray diffraction at the ID27 beamline at ESRF, the P02.2 beamline at PETRA III, and the PSICHÉ beamline at SOLEIL. In all cases, olivine was compressed at ambient temperature up to pressures between 2 and 10 GPa, after which we laser-heated to achieve temperatures ranging between 1400 and 1800 K. Pressure was then slowly increased while maintaining a constant temperature within ±100 K. The sample composition and pressure were monitored from the 2D-diffraction patterns using the software Dioptas (Prescher and Prakapenka 2015). Sample temperatures were measured using spectral radiometry as provided by the beamlines during the experiments. At several points upon compression, pressure increase was stopped to collect multigrain X-ray diffraction data by rotating the DAC over Δω ≈ 60° (depending on diamond height and cell opening) and acquiring X-ray diffraction images every 0.5° rotation increment δω (Fig. 2). At both ESRF-ID27 and PETRA-III-P02.2, laser heating was maintained during multigrain X-ray diffraction data collection, using the standard laser-heating configuration at ESRF, and using a device such as that presented in Bykova et al. (2019) at PETRA-III. At SOLEIL-PSICHÉ the sample was quenched to ambient temperature before multigrain X-ray diffraction data collection. Sample recrystallization could occur during the high-temperature 3D-XRD scans. Note, however, that recrystallization is most expected in deformed grains with high concentrations of internal defects (e.g., Poirier and Guillopé 1979; Urai et al. 1986), and hence parent olivine rather than the newly nucleated wadsleyite grains. In situ 3DXRD measurements at high temperature, nevertheless, offer the great advantage to avoid introducing additional deviatoric stresses upon temperature quenching that could, also, affect the sample (Kavner and Duffy 2001).

At the P02.2 beamline at PETRA III, X-rays were set to a wavelength of 0.2903 Å and focused to 7.6 (horizontal) × 4.4 (vertical) μm. The sample-to-detector distance was 398.7 mm, based on the CeO₂ calibration. Diffraction images were collected using a PerkinElmer XRD 1621 amorphous silicon flat-panel detector with 2048 × 2048 pixels of 200 × 200 μm (Liermann et al. 2015) for 5 s. At the ID27 beamline at ESRF, we used X-rays with a wavelength of 0.3738 Å, focused to 3.2 (horizontal) × 3.0 (vertical) μm. The sample-to-detector distance was 245.54 mm, based on the CeO₂ calibration. Diffraction images were collected using a MAR165 CCD planar detector with 2048 × 2048 pixels of 79 × 79 μm for 6 s. At the PSICHÉ beamline at SOLEIL, we used X-rays with a wavelength of 0.3738 Å focused to about 12 (horizontal) × 10 (vertical) μm. The sample-to-detector distance was 389.75 mm, based on the CeO₂ calibration. Diffraction images were collected employing a PILATUS flat-panel detector with 1475 × 1679 pixels of 172 × 172 μm, and with an acquisition time of 5 s. Note that the diffraction study of large single crystals was allowed by the PILATUS detector that can count every photon received on the panel, with high peaks spread resolution and a small readout time (Eikenberry et al. 2003), and can handle significantly higher peak intensity before saturation.

Table 1 summarizes the experiments that were performed along with the starting materials, pressure-transmitting media, and sample characterization techniques. Three experiments used pure polycrystalline olivine samples as starting materials: LTC_05_01 and LTC_03_02 performed on ESRF-ID27 and P2_01 performed on PETRA-III-P02.2. One experiment started from an olivine single-crystal, Olivine_01 performed at SOLEIL-PSICHÉ.

Diffraction data processing

Post-experiment, ω-rotation multigrain diffraction images were stacked to generate an average diffraction image, which was then further processed using the Rietveld refinement package MAUD (Lutterotti et al. 2014; Wenk et al. 2014) from which we extract the unit-cell parameters of the pressure medium, olivine, and wadsleyite. The pressure was then calculated using thermal equations of state from the literature [Angel et al. (2018) for olivine and Katsura et al. (2009) for wadsleyite] and the EosFit Calculator software (Angel et al. 2014). Pressure and temperature conditions for each multigrain measurement are summarized in Table 2.

The sample microstructural evolution during the transformation is processed using multigrain crystallography (MGC). This method involves acquiring 2D-diffraction images at different ω rotation angles. The data are then processed to: (1) separate the background and the diffraction signals of the pressure medium from that of the larger sample grains; (2) identify individual diffraction spots of the larger sample grains; and (3) determine the sample grain orientation matrices. The detailed procedure has been described in Rosa et al. (2015) and used in Rosa et al. (2016) and Langrand et al. (2017). It relies on open-source softwares from the FABLE-3DXRD package, available online at https://github.com/FABLE-3DXRD,additional tools from the TIMEleSS project, available online at https://github.com/FABLE-3DXRD/TIMEleSS, and thoroughly described in a dedicated manual at http://multigrain.texture.rocks/.

Individual diffraction images are first cleaned to exclude high-intensity single-crystal diffraction spots from the diamond anvils and subtract the background signal. Individual diffraction spots are then located using a high-pass filter, as implemented in the Peaksearch algorithm (Sörensen et al. 2012). The ImageD11 software (Wright 2006) then uses the list of extracted diffraction spots and information on the experimental conditions to compute a list of potential G-vectors to be assigned to the sample grains. ImageD11 does not include full crystallographic information to generate the list of G-vectors and hence generate Miller indices which can be extinct in a given crystal structure. We hence re-evaluate the list of potential G-vectors using our TIMEleSS scripts and the full crystallographic information for olivine and wadsleyite, and concentrate on G-vectors that will actually contribute to diffraction. This additional step decreases the risk of error during grain indexing. At this point, each observed G-vector is assigned to its observed intensity and location in orientation space (2θ, η, and ω angles), and potential h, k, and l Miller indices.

The final indexing step relies on the GrainSpotter software (Schmidt 2014), with the following settings: 100 000 random grain orientation trials, tolerances of 0.08° in 2θ, 4° in η and 3° in ω to assign a measured G-vector to a grain, a minimum number of 15 G-vectors per grain, and a minimum completeness of 30%. This procedure is repeated several times (i.e., 50), removing already assigned G-vectors, to improve the number of indexed grains. The final output consists of a list of grains, along with their crystallographic orientation and the list of the associated diffraction spots.

From the results of the MGC processing, it is possible to extract a relative size for the indexed grains by scaling grain volume to the intensity of their diffraction spots. Then, using the illuminated volume of the sample, the number of indexed grains and their relative volumes, we can provide an estimate of their actual size, as volume or mean radius. This computation is performed using a script from the TIMEleSS tools.

The raw multigrain data, as well as the resulting lists of grains, their orientation and grain size are available online at https://doi.org/10.57745/NZFWP9(Ledoux et al. 2023).

Grain orientations analysis

Indexed grain orientations are shown as either pole figures (PF) (for samples with very few grains, e.g., Fig. 4) or inverse pole figures (IPF) of the compression direction (for samples with a large number of grains, e.g., Fig. 5). Each marker in inverse pole figures represents an indexed orientation and is color-coded according to calculations based on an orientation distribution function (ODF) fitted to the indexing results. Markers in pole figures also represent the orientation of indexed grains but are not color-coded because orientation distributions are not relevant for a low number of grains. For this step, we use the open-source MTEX toolbox for MATLAB (Bachmann et al. 2010).

Figure 4

Transformation microstructures starting from an olivine single crystal. Single-crystal olivine data (a and b), olivine-wadsleyite polycrystal data after partial transformation (c, d, and e). Stacked diffraction patterns (a and c). (100), (010), and (001) pole figures representing the orientations of olivine (b and d) and wadsleyite grains (e). The compression direction is indicated above the pole figures. To better illustrate the difraction spots of the wadsleyite grains, a zoom-in for one image at one omega step of the collection is shown (f).

Figure 5

Transformation microstructures in polycrystalline olivine samples. Inverse pole figures of the compression direction representing grain orientations of olivine (Ol) before transformation, olivine (Ol) and wadsleyite (Wd) during transformation, and wadsleyite (Wd) after transformation from olivine. LTC_05_01, LTC_03_02, and P2_01 are different experiments. Pressure and temperature conditions of the measurements are indicated below each inverse pole figure along with N, the number of indexed grains. The fundamental space and axes defining the inverse pole figure are shown on the right. The color scale applied to the markers is based on a probability calculated from the orientation distribution function fitted to the sample and is expressed in multiple of a random distribution (m.r.d.). It should be noted that, in all the experiments in this figure, MGC data collections were performed in situ at high P-T conditions during laser-heating.

MTEX is also used to test the effect of a coherent transformation mechanism from olivine to wadsleyite. For these simulations, we start from the indexed olivine grain orientations prior to the transformation and compute the daughter wadsleyite orientations resulting from a strict coherent transformation. We test for both orientation relationships suggested by Smyth et al. (2012): Type I with [001]_ol || [010]_wd and (100)_ol || {101}_wd, and Type II with [001]_ol || [100]_wd and (100)_ol || {021}_wd. The simulated daughter wadsleyite orientations are also plotted as IPF, using the same procedure described above.

Transformation microstructures in olivine single crystal

Sample in run Olivine_01 was an olivine single crystal at 6.3 GPa and 300 K. It was laser-heated to temperatures of 1700–1800 K and compressed at 1700–1800 K over 100 min to drive a partial conversion to wadsleyite. Multigrain crystallography data was collected on the olivine single-crystal prior to the pressure increase, and at 16.1 GPa and 300 K, after the partial transformation to wadsleyite and quench in temperature (Fig. 4).

Before the transformation (Figs. 4a and 4b), the sample is an olivine single crystal, as shown by both the diffraction pattern and the single-grain orientation deduced from the multigrain crystallography processing. The diffraction image shows intense and slightly deformed diffraction peaks, indicating that the crystal is large and is submitted to some stress. The multigrain crystallography processing locates a single grain, with an estimated equivalent radius (i.e., the radius of a sphere with an identical volume to that of the grain) of 6.1 μm.

Later, upon compression at high temperature (Figs. 4c–4f) the sample contains both olivine and wadsleyite. Olivine diffraction peaks are still intense (Fig. 4), indicating grains of larger size, but also appear elongated along the diffraction rings, suggesting that the olivine portion of the sample is subdividing into different orientation domains. The multigrain crystallography processing identifies 25 grains of olivine with a mean radius of 0.8 μm. The orientations of these grains do not show any particular alignment with specific sample directions, but are different from the orientation of the original single crystal, before partial transformation to wadsleyite (Fig. 4b). On the stack of all ω diffraction patterns, wadsleyite appears as small diffraction peaks forming a nearly continuous diffraction ring, suggesting that wadsleyite grains are small. On a single image of the collection, however, the diffraction spots of wadsleyite are distinct (Fig. 4f). The multigrain crystallography processing locates 72 daughter orientations of wadsleyite, with a mean radius of 0.8 μm. Most of these grains show a cluster of orientation with their [100] crystallographic axes in the vicinity of the compression direction (Fig. 4e).

Transformation microstructures in polycrystalline olivine

Samples in runs LTC_05_01, LTC_03_02, and P2_01 were pure polycrystalline olivine. All were compressed at temperatures ranging between 1400 and 1730 K and converted to wadsleyite. Figure 5 presents the orientations of the grains indexed by MGC in these three experiments and at different steps of the transformation.

In experiment LTC_05_01, the transformation from olivine to wadsleyite is observed between 1400 K–14.9 GPa and 1730 K–18.7 GPa. Before transformation (collection s26), MGC indexes 55 olivine grains with a mean grain radius of 0.5 μm and no lattice preferred orientation. After the transformation (collection s27), MGC locates 114 grains of wadsleyite with a mean grains radius of 0.6 μm. We observe significant crystal preferred orientations, with the [100] crystallographic axes of wadsleyite aligned at a low angle of the compression direction.

In experiment LTC_03_02, the transformation from olivine to wadsleyite is observed at 1700 K and between 17.0 and 18.8 GPa. Before the transformation (collection s07), MGC indexes 180 grains of olivine with a mean grain radius of 0.5 μm. The olivine grain orientations form a girdle between the (100) and (010) poles of the inverse pole figure. After the transformation (collection s08), 188 grains of wadsleyite are indexed with a mean grain radius of 0.5 μm and a crystal preferred orientation with the maximum orientation probability 30° away from (100).

In experiment P2_01, the transformation from olivine to wadsleyite is observed between 1550 K–20.2 GPa and 1450 K–12.3 GPa. Before the transformation (collection s06), MGC indexes 84 olivine grains with a mean grain radius of 1.0 μm and no crystal preferred orientations. In this experiment, pressure decreased at the olivine to wadsleyite conversion, with data collected while olivine and wadsleyite coexist (collection s07) leading to 83 olivine grains with a mean grain radius of 0.7 μm and no crystal preferred orientation and 30 wadsleyite grains with a mean grain radius of 0.7 μm and orientations concentrated between the (010) and (001) poles of the inverse pole figure. After full conversion to wadsleyite (collection s08), we index 144 wadsleyite grains with a mean grain radius of 0.7 μm. The crystal preferred orientation is weak but shows several maxima at (001), (010), and (001) in the inverse pole figure.

Grain size evolution

Two tendencies are observed regarding grain size evolution in our experiments: (1) in experiments Olivine_01 and P2_01, the mean sample grain size decreases during the phase transition from olivine to wadsleyite, while (2) in experiments LTC_03_02 and LTC_05_01, the grain size does not significantly change during the phase transition.

The grains size evolution in experiments Olivine_01 and P2_01 is illustrated in Figure 6, based on the example of experiment P2_01 for which data with coexisting olivine and wadsleyite is available. Before the transformation (data set s06), the olivine grain radii vary between 0.5 and 2.2 μm. During the transformation, the olivine grain size decreases, with grain radii between 0.4 and 1.6 μm (data set s07). Coexisting wadsleyite in the same measurement shows grain radii between 0.4 and 1.1 μm. After the transformation (data set s08), the sample is made of wadsleyite grains with grain radii ranging between 0.2 and 1.8 μm and an average radius of 0.7 μm.

Figure 6

Distribution of estimated grain equivalent radii in experiment P2_01 while the sample is pure olivine (collection s06), a mixed olivine-wadsleyite phase (collection s07), and fully transformed to wadsleyite (collection s08). The green and blue bars correspond to olivine and wadsleyite grains size, respectively. n is the number of grains for each phase. M, m, and Av are the maximum, minimum, and average grain radii, respectively. MD is the grain radius for which distribution is maximal. All are expressed in micrometers.

Regarding experiments LTC_03_02 and LTC_05_01, for which grain size does not reduce upon the phase transformation to wadsleyite, we can notice that the average grain radius in olivine before transformation is 0.5 μm. In all experiments, the mean radius of wadsleyite grains after full transformation is above 0.5 μm. We can hence conclude that, in experiments LTC_03_02 and LTC_05_01, olivine grain size before transformation is already close to that of the newly formed wadsleyite grains. The phase transformation hence does not result in grain size reduction in these experiments.

Overall, the grain sizes in our experiments are small with an estimated mean grain size equal to or smaller than 1 μm in equivalent radius (except for the starting single crystal of Olivine_01). The differences in mean wadsleyite grain sizes between experiments can be explained by the grain size in the starting material. Nevertheless, wadsleyite grain sizes are somewhat similar in all experiments and we do not see a clear effect of the pressure medium. More investigations will be needed, however, to conclude on the relative effects of stress, overpressure, and pressure media on wadsleyite grain sizes after transformation. In all cases, large olivine crystals tend to disappear during the transformation, with olivine grain radii of about 0.8 μm in equivalent radius in coexisting olivine and wadsleyite. The wadsleyite grains past phase transformation have mean grain radii between 0.5 and 0.8 μm.

Olivine-wadsleyite transformation mechanism

We observe lattice preferred orientations in wadsleyite after transformation, with [100] crystallographic axis parallel or at 30–40° of the compression direction in experiments LTC_05_01 and LTC_03_02 (Fig. 5). Starting from an olivine single-crystal (Olivine_01, Fig. 4), wadsleyite after transformation also shows a similar LPO, although weaker than in experiments LTC_05_01 and LTC_03_02.

Experiment P2_01, starting from polycrystalline olivine and for which measurements in the mixed olivine-wadsleyite phase are available, shows a more complex behavior. The first wadsleyite grains nucleate with orientations between the (010) and (001) poles of the IPF of the compression direction (Fig. 5 s07). After the transformation (collection s08), the orientation distribution spreads, with clusters of orientations near the (010), (001), and (100) poles of the IPF, which could be a transition between the first nucleation texture (collection s07) and a texture with the [100] crystallographic axis parallel to compression, as in the three other experiments.

A coherent transformation implies a crystallographic orientation relationship between the parent-phase, olivine, and the daughter-phase, wadsleyite. Smyth et al. (2012) report two topotactic relationships for the olivine-wadsleyite coherent transformation, based on observations in transmission electron microscopy of post-mortem products of the transformation. Type I orientation relationships lead to (100) olivine || {101} wadsleyite and [001] olivine || [010] wadsleyite. Type II orientation relationships lead to {021} wadsleyite || (100) olivine and [100] wadsleyite || [001] olivine. Both Type I and Type II orientation relationships are tested in Figure 7 for experiments LTC_05_01 and LTC_03_02. The results are clear: the orientations of wadsleyite grains observed in the experiment do not correspond to the predictions of either Type I or Type II orientation relationships.

Figure 7

Coherent transformation simulations (b, d, f, and h) and comparison with the experimental results (a, c, e, and g), for experiments LTC_05_01 (a, b, c, and d), and LTC_03_02 (e, f, g, and h). For each experiment, the olivine grain orientations before transformation (a and e) are used to compute the daughter wadsleyite orientations. The computation is based on the two topotactic relationships given by Smyth et al. (2012), indicated below the corresponding inverse pole figures. Panels c and g are experimental wadsleyite grains orientations just after the transformation.

In P2_01, no clear LPO is observed in the olivine grains before the transformation or in the relict olivine grains coexisting with wadsleyite during transformation. Based on our simulations (Fig. 7), coherent transformations of polycrystalline olivine with no LPO should produce daughter-wadsleyite with no LPO. However, the first wadsleyite grains observed during the transformation show a relatively strong LPO along a girdle between the (010) and (001) poles of the IPF of the compression direction. Hence a coherent transformation cannot explain the transformation texture (nor the post-transformation texture) of the newly formed wadsleyite in this experiment.

Experiments on polycrystalline olivine hence show that orientation relationships observed at the local scale do not apply at the polycrystal scale. As discussed in Smyth et al. (2012), incoherent grains of wadsleyite dominate the sample microstructure. This result is confirmed with experiment Olivine_01, starting from a single crystal, in which multiple domains of wadsleyite and olivine are formed during transformation, with no obvious orientation relation to the original single-crystal. As such, the inheritance of crystallographic preferred orientations in the olivine-wadsleyite transformation is unlikely to be significant at the 410 km depth discontinuity in the Earth mantle, with a dominant incoherent nucleation of wadsleyite. LPO in wadsleyite during or after transformation is most likely related to oriented growth due to stress or plastic deformation following the phase transformation.

Interestingly, the grain size observed in wadsleyite in our LH-DAC experiments is close to those observed in multi-anvil press experiments at 1400–1800 K with durations ranging from tens of minutes to a few hours (Nishihara et al. 2006; Demouchy et al. 2011; Mohiuddin and Karato 2018). This observation supports that transformation mechanisms in our LH-DAC experiments should not be significantly different than that deduced from large-volume press experiments.

Effect of grain size on the transformation mechanism

The study of Kerschhofer et al. (1996) shows that the grain size of the transforming olivine influences the olivine-wadsleyite transformation mechanism with: (1) a grain-boundary nucleation of wadsleyite, operating in both a large single crystal and a fine grained matrix, and (2) an intracrystalline nucleation of wadsleyite inside a large olivine single crystal only. None of these mechanisms are reported to produce strict orientation relationships between olivine and wadsleyite.

In our polycrystalline experiments, the mean grain size of the parent-olivine is up to 1.0 μm in equivalent radius and hence is smaller than that of the fine matrix of the experiments of Kerschhofer et al. (1996), where the grain-boundary nucleation is dominant. In experiment Olivine_01 the single-crystal olivine sample is shown to subdivide into multiple domains of both olivine and wadsleyite as the phase transition proceeds (e.g., Fig. 4), leading to an average grain radius of 0.8 μm both in olivine and wadsleyite. According to Kerschhofer et al. (1996), the phase transition, in this case, should also be controlled by a dominant grain-boundary nucleation mechanism.

Mohiuddin and Karato (2018) also found wadsleyite grains of small size (2.8–4.2 μm in diameter) and proposed dominant grain-boundary nucleation of wadsleyite, but at the condition that the transformation over-pressure is below 3 GPa. As over-pressure increases, they showed that intra-crystalline nucleation makes a significant contribution to the volume fraction transformed. In our experiments Olivine_01, LTC_05_01, and LTC_03_02, the transformation over-pressure was less or equal to 3 GPa, and so should be dominated by grain-boundary nucleation of wadsleyite, in agreement with the proposition of Kerschhofer et al. (1996). In experiment P2_01, although grain size is similar to that of the other samples, the over-pressure is larger than 3 GPa. Thus, based on the results of Mohiuddin and Karato (2018), a greater contribution of the intra-crystalline mechanism for wadsleyite nucleation could be expected in this particular experiment.

Effect of stress on the phase transformation

Transformation mechanisms in the olivine system are known to be sensitive to parameters such as stress, temperature, and over-pressure (e.g., Burnley and Green 1989; Burnley et al. 1995; Kerschhofer et al. 1996, 1998; Smyth et al. 2012, and references therein). Early diamond-anvil cell experiments provided inconsistent results, probably due to the lack of pressure-transmitting media and high stresses (Burnley et al. 1995).

Our experiments are performed at much higher temperatures, thanks to laser heating, and use different pressure media, MgO, alumina, and KCl, all with different strengths. They lead to consistent results: the wadsleyite microstructures after transformation from olivine are dominated by an incoherent mechanism, whatever the pressure medium or over-pressure prior to the transformation, with no effect of the olivine starting texture or grain sizes, and consistent grain sizes in wadsleyite past transformation (e.g., Table 2). This is consistent with observations in low-stress and small grain sizes experiments in large-volume press experiments (e.g., Burnley and Green 1989; Kerschhofer et al. 1996; Smyth et al. 2012). A more precise analysis of the effect of stress on the phase transformation mechanism would require a proper estimate of stress in the olivine and wadsleyite grains. Grain-to-grain stress estimate can be obtained using 3D-XRD (e.g., Oddershede et al. 2010; Chandler et al. 2021a) but requires careful calibration for relevant measurements. We did attempt to extract grain-level stress distributions from our data sets but did not obtain reliable results. This will have to be the topic of future investigations.

Over-pressure, stress, and subsequent deformation may have affected details of grain orientations in wadsleyite past transformation from olivine (e.g., Figs. 4 and 5), but the fine analysis of these features goes beyond the goals of this paper.