Generating porosity during olivine carbonation via dissolution channels and expansion cracks

The olivine carbonation reaction, in which carbon dioxide is chemically incorporated to form carbonate, is central to the emerging carbon sequestration method using ultramafic rocks. The rate of this retrograde metamorphic reaction is controlled, in part, by the available reactive surface area: as the solid volume increases during carbonation, the feasibility of 10 this method ultimately depends on the maintenance of porosity and the creation of new reactive surfaces. We conducted insitu dynamic x-ray microtomography and nanotomography experiments to image and quantify the porosity generation during olivine carbonation. We designed a sample setup that included a thick-walled cup (made of porous olivine aggregates with a mean grain size of either ~5 or ~80 μm) filled with loose olivine sands with grain sizes of 100-500 μm. The whole sample assembly was reacted with a NaHCO3 aqueous solution at 200 °C, under a constant confining pressure of 13 MPa and a pore 15 pressure of 10 MPa. Using synchrotron-based X-ray microtomography, the 3-dimensional (3-D) pore structure evolution of the carbonating olivine cup was documented until the olivine aggregates became disintegrated. The dynamic microtomography data show a volume reduction in olivine at the beginning of the reaction, indicating a vigorous dissolution process consistent with the disequilibrium reaction kinetics. In the olivine cup with a grain size of ~80 μm (coarse-grained cup), dissolution fractures developed within 30 hours, before any precipitation was observed. In the experiment with the olivine cup of ~5 μm 20 mean grain size (fine-grained cup), idiomorphic magnesite crystals were observed on the surface of the olivine sands. The magnesite shows a near constant growth throughout the experiment, suggesting that the reaction is self-sustained. Large fractures were generated as reaction proceeds and eventually disintegrate the aggregate after 140 hours. Detailed analysis show that these are expansion cracks caused by the volume mismatch between the expanding interior and the nearly constant surface. Nanotomography images of the reacted olivine cup reveal pervasive etch-pits and worm-holes in the olivine grains. We 25 interpret this perforation of the solids to provide continuous fluid access, which is likely key to the complete carbonation observed in nature. Reactions proceeding through the formation of nanoto micron-scale dissolution channels provide a viable microscale mechanism in carbon sequestration practices. For the natural peridotite carbonation, a coupled-mechanism of dissolution and reaction-induced fracturing should account for the observed self-sustainability of the reaction.

Abstract.The olivine carbonation reaction, in which carbon dioxide is chemically incorporated to form carbonate, is central to the emerging carbon sequestration method using ultramafic rocks.The rate of this retrograde metamorphic reaction is controlled, in part, by the available reactive surface area: as the solid volume increases during carbonation, the feasibility of this method ultimately depends on the maintenance of porosity and the creation of new reactive surfaces.We conducted in situ dynamic X-ray microtomography and nanotomography experiments to image and quantify the porosity generation during olivine carbonation.We designed a sample setup that included a thick-walled cup (made of porous olivine aggregates with a mean grain size of either ∼ 5 or ∼ 80 µm) filled with loose olivine sands with grain sizes of 100-500 µm.The whole sample assembly was reacted with a NaHCO 3 aqueous solution at 200 • C, under a constant confining pressure of 13 MPa and a pore pressure of 10 MPa.Using synchrotron-based X-ray microtomography, the three-dimensional (3-D) pore structure evolution of the carbonating olivine cup was documented until the olivine aggregates became disintegrated.The dynamic microtomography data show a volume reduction in olivine at the beginning of the reaction, indicating a vigorous dissolution process consistent with the disequilibrium reaction kinetics.In the olivine cup with a grain size of ∼ 80 µm (coarsegrained cup), dissolution planes developed within 30 h, before any precipitation was observed.In the experiment with the olivine cup of ∼ 5 µm mean grain size (fine-grained cup), idiomorphic magnesite crystals were observed on the surface of the olivine sands.The magnesite shows a near-constant growth throughout the experiment, suggesting that the reac-tion is self-sustained.Large fractures were generated as the reaction proceeded and eventually disintegrated the aggregate after 140 h.Detailed analysis show that these are expansion cracks caused by the volume mismatch in the cup walls, between the expanding interior and the near-surface which keeps a nearly constant volume.Nanotomography images of the reacted olivine cup reveal pervasive etch pits and wormholes in the olivine grains.We interpret this perforation of the solids to provide continuous fluid access, which is likely key to the complete carbonation observed in nature.Reactions proceeding through the formation of nano-to micronscale dissolution channels provide a viable microscale mechanism in carbon sequestration practices.For the natural peridotite carbonation, a coupled mechanism of dissolution and reaction-induced fracturing should account for the observed self-sustainability of the reaction.

Introduction
Mantle peridotites are exposed widely on the Earth's surface in tectonic settings such as mid-ocean ridges, subduction zones and ophiolites (Escartín et al., 1997;Fryer et al., 1995).Peridotite is mainly composed of olivine which is unstable at temperatures below 700 • C in the presence of water (Evans, 1977) and below 500 • C in the presence of CO 2 -rich fluids (Johannes, 1969).The transformation of olivine to serpentine and carbonates due to fluid-rock interaction is extensively observed in peridotite outcrops (e.g., Beinlich et al., 2012;Falk and Kelemen, 2015;Hansen et al., 2005).Rock deformation experiments have demonstrated that fluid alteration to Published by Copernicus Publications on behalf of the European Geosciences Union.
T. Xing et al.: Generating porosity during olivine carbonation peridotite can strongly affect the strength and tectonics of the oceanic lithosphere (Deschamps et al., 2013;Moore et al., 1996).Therefore, study of the olivine-fluid interaction is of great importance for understanding the alteration process of peridotite in a variety of tectonic regions.General peridotite alteration reactions can be formulated as follows (Hansen et al., 2005;Kelemen and Matter, 2008 (R4) Although peridotite weathering reactions occur widely in nature, the rate of olivine carbonation at subsurface conditions is debated.Since the retrograde metamorphic reactions are kinetically fast, the extent of transformation is limited by fluid supply which depends on the accessible fluid pathways.As the hydration and carbonation of olivine results in an up to ∼ 44 % increase in solid molar volume (Goff and Lackner, 1998;Hansen et al., 2005;Kelemen and Matter, 2008), carbonation of olivine is generally assumed to be selflimiting: the reaction products would gradually fill up the pore space and lead to a decrease in the porosity (Emmanuel and Berkowitz, 2006;Hövelmann et al., 2012), which in turn lowers permeability and reduces fluid supply.This negative feedback would ultimately force the alteration to cease.However, naturally occurring completely carbonated peridotites are evidence that these limitations can be overcome.For instance, listvenite is the natural completely carbonated product of peridotite, which is composed of magnesite, quartz and trace minerals (Beinlich et al., 2012;Nasir et al., 2007).This creates a conundrum of how the large extent carbonation can be achieved with the potential self-limitation of reducing fluid pathways.
In order to explain this discrepancy between the theory and the observation, numerous studies have been conducted aiming to find a mechanism to maintain the access for reacting fluid during olivine alteration reactions.In 1985, Macdonald and Fyfe examined naturally altered peridotite and proposed that the large volume change associated with the reaction could generate high local stresses and strains, which would cause episodic cracking.This idea has then been applied to olivine carbonation by Kelemen and Matter (2008), who proposed a positive feedback loop where fractures could be generated during the volume-expanding reaction, porosity and permeability can be maintained or even increased, which in turn would accelerate the carbonation processes (Rudge et al., 2010).In 2011, Kelemen et al. showed that in natural peridotites cross-cutting hierarchical fracture networks filled by synkinematic carbonate and quartz veins extend to microscopic scales.These cross-cutting networks indicate coeval carbonate crystallization and fracturing.Several studies also showed that the forces generated by the volume increase should be enough to fracture peridotite (Iyer et al., 2008;Jamtveit et al., 2009Jamtveit et al., , 2011;;Ulven et al., 2014).
While reaction-induced fracturing is accepted as a way to maintain fluid access, the mechanical details of the process are poorly understood.As for the mechanism that generates stresses, "crystallization pressure" (also termed "force of crystallization") has been proposed (e.g., Scherer, 2004;Weyl, 1959;Winkler and Singer, 1972).In this model, the precipitation/crystallization of reaction products exert pressure around the growing crystals, and fracturing takes place when that pressure exceeds the local minimum principal stress (Kelemen and Hirth, 2012).Salt crystallization (Scherer, 2004) is a common example where high crystallization forces due to the nucleation of precipitates in pore space cause samples to "burst from the inside" (see Fig. 1a).However, studies have shown that the crystallization force is low in the olivine carbonation system (e.g., van Noort et al., 2017).Because of the lack of experimental evidence of crystallization forces during olivine alteration, Zhu et al. (2016) proposed an "expansion cracking" mechanism as an alternative model after successfully producing reactioninduced fractures in an in situ synchrotron X-ray microtomography study.In the expansion cracking model, tensile stresses are generated due to the volume mismatch between regions with different precipitation rates, leading to cracks forming in regions that expand slower than their surroundings (see Fig. 1b).
Beyond fracturing, dissolution has been recognized as an important part of the olivine alteration process (e.g., Velbel, 2009;Velbel and Ranck, 2008;Wilson and Jones, 1983) and proposed as a mechanism to explain the observed complete carbonation of peridotite.In 1978, Grandstaff showed that dissolution could significantly increase the surface area through etch pitting.Wilson (2004) suggested that the weathering of olivine is controlled by etch pitting and channel formation due to preferential dissolution, which assists the migration of fluid and promotes further reaction.Andreani et al. (2009) suggested that permeability may be maintained during peridotite carbonation by the development of preferential flow zones.Lisabeth et al. (2017a, b) observed relevant structures in dunite samples that have been reacted under controlled stress conditions and interpreted them as a pattern of secondary porosity bands formed by dissolution coupled to locally intensified compressional stresses.
Previous investigations of olivine carbonation were largely based on the interpretation of naturally deformed samples (e.g., Macdonald and Fyfe, 1985), thermodynamic modeling (e.g., Kelemen and Hirth, 2012) or comparison with reaction systems other than olivine (e.g., leucite to analcime in Jamtveit et al., 2009).While these approaches led to signifi-  (Scherer, 2004) is a typical example of the crystallization pressure-induced fracturing.The fractures first appear at areas where precipitation is most concentrated and propagate outwards.(b) The surface cracking model describes the development of fractures as a result of a contrast in expansion which causes stretching at the surface.A difference in the precipitation rate between the periphery and interior of the sample causes them to expand at different rates with the interior expanding faster than the surface.This builds up the tensile stress at the surface that fractures the sample and leads to a development of a polygonal fracture network.The fractures propagate from the surface inwards.
cant advancements, there are limitations to the understanding of the mechanisms responsible for porosity generation during olivine carbonation that these approaches can provide.The history of natural fault rocks is inevitably complex, and thermodynamic arguments and numerical models can only indicate a potential, while the actual progress of chemical reactions is strongly affected by interfacial structures, which vary considerably in different mineral systems.Thus, it is critical to complement such studies with laboratory experiments on olivine carbonation.
Synchrotron-based X-ray tomography is an advanced nondestructive method to capture three-dimensional images of materials.Where processes affecting these materials are followed through time, a four-dimensional (three spatial dimensions plus time) dataset is captured.By using X-ray transparent reaction cells (Fusseis et al., 2014b), the technique enables the investigation of fluid-rock interaction at controlled and geologically relevant conditions.We examined the carbonation process of olivine on the basis of four-dimensional images acquired by X-ray microtomographic imaging with synchrotron radiation at the Advanced Photon Source.Zhu et al. (2016) hypothesized that large grains would be preferred sites for precipitation of new crystals, and the contrast in the grain size produced the volume mismatch due to a preferred precipitation on the larger grains and led to the fracturing of the sample.To further test the hypothesis, in the current study, we conducted a new experiment using an olivine aggregate with larger grain size (80-100 µm) compared to the previous experiment reported by Zhu et al. (2016;0-20 µm).We also performed advanced 3-D analyses and quantification of the micro-and nanotomography data obtained by Zhu et al. (2016).In an advancement of the results presented by Zhu et al. (2016), here we present direct evidence for the coupled mechanisms of dissolution and precipitation-driven fracturing during olivine carbonation and demonstrate their importance in sustaining the reaction progress at different spatial and temporal scales.We further show direct evidence of how reaction-induced fracturing operates, i.e., how stress is generated through volume-increasing reactions.A better understanding of olivine carbonation directly applies to the geological sequestration of CO 2 (Gislason et al., 2010;Mani et al., 2008).The principle of in situ carbon mineralization is the conversion of silicate and hydroxide minerals to form carbonate minerals as a stable sink for CO 2 (Power et al., 2013).Peridotite, because of its wide occurrence and high reactivity, is considered one of the best potential feedstocks for CO 2 mineralization (Andreani et al., 2009;Beinlich and Austrheim, 2012); the estimated rate of CO 2 consumption peridotite carbonation could be as high as 2 × 10 9 t km −3 per year (Kelemen and Matter, 2008).As the dominant constituent of peridotite, olivine becomes the most important mineral for CO 2 mineralization.Our study provides new insights into carbon sequestration using ultramafic rocks, and our findings on the mechanism of fracture generation during olivine carbonation could provide guidance to industrial applications.
2 Experimental setup

Sample configuration
In Zhu et al. (2016), the contrast in grain size between the loose grains and the cup wall aggregate is hypothesized as the cause of the non-uniform precipitation which is crucial to the generation of fractures in the experiment.Here, we use coarse-grained olivine aggregate in the cup wall to reduce the www.solid-earth.net/9/879/2018/Solid Earth, 9, 879-896, 2018 contrast in grain size between the cup wall and the fillings and further test their hypothesis.
The sample assembly consists of a millimeter-sized synthesized San Carlos olivine cup, filled with loose San Carlos olivine sand (grain size 100-500 µm; see Fig. 2a) and then covered by a lid made of the same material as the cup.Both the olivine cup and its lid are taken from a sintered pellet made from pulverized San Carlos olivine with a grain size of 80-100 µm.This coarse-grained olivine cup is referred to as large grain cup (LGC) in the following discussion.Data obtained from the LGC experiment will be compared to the results of an olivine carbonation experiment conducted by Zhu et al. (2016) at the same experimental conditions, on a fine-grained (grain sizes between 0 and 20 µm) cup.The finegrained olivine cup is referred to as the small grain olivine cup (SGC) in the following discussion.
The cup, which was fabricated by heat pressing in a procedure described in Zhu et al. (2016), has inner and outer diameters of 1 and 1.8 mm, respectively, with a resulting wall thickness of ∼ 0.4 mm.The initial porosity of the cup is ∼ 10 %.Both LGC and SGC samples were sintered for 4 h at 1400 • C, the LGC sample came out weaker (less cohesive) compared to the SGC sample, and the initial porosity of LGC is slightly higher than that of SGC.
The loose grains inside the cup allowed the inspection of magnesite growth on free olivine surfaces.The sample assemblies (i.e., olivine cup and loose grains) were jacketed and loaded into an X-ray transparent pressure cell (Fig. 3).A confining pressure of 13 MPa and a pore fluid of NaHCO 3 aqueous solution (1.5 mol L −1 ) at 10 MPa were applied to the sample.The pore pressure downstream is regulated by a pressure regulator while the pressure upstream is controlled by a syringe pump to form a semi-open system.Both upstream and downstream pore pressures were maintained at 10 MPa throughout the experiment.A small axial load is applied independently from the confining pressure by locking the pis-ton at a fixed position.The pressure cell was then heated to 200 • C to initiate the reaction.These conditions were kept constant during the entire microtomography experiment.The reaction of the LGC sample was stopped at 36 h after considerable secondary porosity generation was observed.No magnesite precipitation was detected in LGC.The SGC sample reacted for 7 days until intense fracturing was observed.Significant magnesite precipitation was detected in SGC (Zhu et al., 2016).

Micro-and nanotomography
Third-generation synchrotron facilities produce electromagnetic radiation bright enough to allow rapid imaging even inside experimental vessels, thereby enabling studies of dynamic processes ranging over periods from seconds to days, while acquiring individual three-dimensional (3-D) datasets in fractions of a second.Synchrotron X-ray microtomography has therefore become one of the most powerful tools in structural geology and rock mechanics studies (see Fusseis et al., 2014a for a review andBedford et al., 2017 for a recent application).
In this experiment, synchrotron-based X-ray absorption microtomography has been used to record the dynamic carbonation of olivine in four dimensions.We used an X-ray transparent cell (Fusseis et al., 2014b), mounted in the upstream experimental station at beamline 2BM of the Advanced Photon Source of the Argonne National Laboratory, 25 m from the source.There, a polychromatic beam filtered by 1 mm aluminum, 15 mm silicon and 8 mm borosilicate glass yielded a photon flux with an energy peak at 65 keV (Zhu et al., 2016).A Cooke pco.edge sCMOS camera with 2560 × 2160 pixels (pixel size 6.5 × 6.5 µm 2 ) was used in a flying scan mode.The sample-detector distance was 300 mm, which introduced a clear phase contrast signal to the data (Cloetens et al., 1996).The camera recorded projections from a 10 µm thick LuAG:Ce single crystal scintillator, magnified through a 10× Mitutoyo long-working distance lens yielding a pixel size of 0.65 µm.Projections were collected with an exposure time of 50 ms while the sample was rotated over 180 • with 1.2 • s −1 .A total of 1500 projections were collected in 150 s.For the LGC sample, 115 3-D microtomographic datasets were acquired over 36 h, together forming a 4-D dataset, with time as the fourth dimension.For the SGC sample, 379 datasets were acquired over 7 days.From these 379 datasets, 19 were chosen for further detailed quantitative analysis.All acquired microtomographic data were reconstructed using the code Tomopy (Gürsoy et al., 2014) into stacks of 2160 images each, with dimensions of 2560 × 2560 pixels per image.Each of these image stacks contains a three-dimensional representation of the sample mapped onto a 32 bit image, with the grey values reflecting the local absorption of X-rays (Fusseis et al., 2014a).Where the refractive indices change in the sample, i.e., on edges, this absorption signal is locally overlain by a phase contrast signal (Cloetens et al., 1996).The time series dataset covers the entire duration of the experiment.
After the in situ acquisition of the microtomography images, a fragment of the cup wall from the SGC was taken to conduct nanoscale imaging.Nanotomography was conducted using a transmission X-ray microscope (TXM) at the beamline 32-ID of the Advanced Photon Source of the Argonne National Laboratory.A monochromatic beam of Xray with an energy of 8 keV was used.An X-ray objective lens corresponding to a Fresnel zone plate with 60 nm outermost zone width was used to magnify radiographs onto a detection system assembly comprising a LuAG scintillator, a Zeiss 5× optical microscope objective lens and an Andor Neo sCMOS camera.Nanotomography yields a pixel size of ∼ 60 nm after binning.

Image processing procedures
Zhu et al. ( 2016) conducted preliminary analyses and measurements on two-dimensional (2-D) image slices (see Fig. 2b) of SGC through the 3-D microtomography datasets acquired.Here, we present the results of a true 3-D volume quantification of the microstructural changes in the SGC sample (i.e., spatiotemporal changes in grain and pore volumes) and compare them with the results obtained from LGC sample in this study.
In both the SGC and the LGC samples, a subregion that included both the cup wall near-surface as well as the cup wall's interior was chosen for detailed inspection from all datasets.For each sample, two subvolumes with the dimension of 400 × 400 × 400 voxels were cropped out from the region of interest in the cup wall (see Fig. 4, subvolumes 1 and 2 from SGC and subvolumes 3 and 4 from LGC).In the SGC sample, subvolume 2 was further cropped to a volume of 247 × 400 × 400 voxels (160.55 × 260 × 260 µm 3 ) to eliminate the boundary of the cup wall.
Image segmentation is the separation and extraction of phases of interests from the 3-D datasets for further analysis and quantification.A large range of segmentation algorithms exists (e.g., Kaestner et al., 2008).In global binary thresholding, images are segmented by identifying the grey value range representing a phase and assigning all voxels within that range a single value (usually 1) while all other voxels are classified as matrix (and assigned a different single value, usually 0) (Heilbronner and Barrett, 2014).Global binarization was conducted in Avizo Fire 8 to isolate pores from solids.
In the SGC sample, at the given spatial resolution, we could not resolve the new crystals that precipitated within the cup wall and only the pore space was segmented there.Pixels with grey values that fall in the range (−0.00031, −0.000077) were assigned to pore space.We used the segmented data to quantify the change in the spatial distribution of pores during the experiment.Each subvolume was further divided into smaller cubes (side lengths ∼ 26 µm) in which the average porosity is calculated to examine where changes in porosity occurred.In the LGC sample, grey values that fall in the range (−0.00031, −0.000095) were assigned to pore space.
In the nanotomography data, the grey value range (2.97672 × 10 −9 , 0.13161) was assigned to pore space.In these data, the olivine is represented by grey values in the range (0.30846, 1.3161).Voxels with intermediate grey values (0.13161, 0.30846) were assigned to reaction precipitates (e.g., magnesite).

Data analysis and results
We have observed the development of secondary porosity during the reaction in both the LGC and SGC experiments (Fig. 5).Detailed examination has further revealed that the porosity generation in LGC is significantly different from that in SGC.

Dissolution in the LGC sample
As the reaction proceeded, olivine grains in the LGC sample shrank in size and secondary pores were produced (Fig. 5a).This suggests that dissolution dominated the carbonation reaction in LGC, with little precipitation detected.The secondary pore space formed first at the center of the cup wall and grew outwards (Fig. 5a), which is opposite to that in SGC where fractures grew from the surface of the cup wall inwards (Fig. 5b).
Planar fractures formed within 36 h of reaction in LGC (Fig. 6), and the experiment was stopped shortly after the formation of these fractures.Because of the lack of precipitation, these planar fractures could not be induced by the non-uniform volume expansion as in the SGC sample (Zhu et al., 2016).Under a constant confining pressure, volume reduction in olivine grains (i.e., dissolution) likely shortened the LGC sample length as the reaction proceeded.Because the axial piston was kept at a fixed position during the experiment, this shortening in sample length resulted a decrease in axial stress.Because the LGC sample is mechanically weak (less cohesion), even though the reduction in axial stress is small, it could be sufficient to cause fracture in LGC in the manner of dilation bands under triaxial extension (e.g., Zhu et al., 1997).Detailed examination of the 3-D images revealed the disappearance of small grains along the plane which is clear evidence of dissolution.Thus, we refer to these planar cracks as dissolution-assisted fractures under triaxial extension.The dissolution-assisted fractures were not observed in the SGC sample because it is much stronger due to its fine grain size (e.g., Eberhardt et al., 1999;Singh, 1988).The triaxial extension stress condition would be no longer present once precipitation started (after ∼ 36 h) and sample volume expansion took place.

Lack of precipitation in the LGC sample
Evidence from direct observation and quantitative analysis has shown that the LGC sample has a lack of precipitation.
Within the duration of the reaction, hardly any precipitation was observed in the cup wall.In the microtomographic images, the surface layer of the loose olivine grains remained free of precipitates.Both sides of the cup wall remained straight and showed no spalling due to precipitation-caused non-uniform stretching.The sample did not experience any expansion as seen in the SGC sample.
Apart from the direct observation, grey value histograms of the data also show evidence of lacking precipitation in  The boarder color of each section corresponds to color code given in Fig. 2.
the LGC sample.The grey value histograms of 4-D microtomography data evolve systematically during in situ experiments, which can be utilized in the evaluation of the reaction progress (Fusseis et al., 2012).Systematic analysis of the his-tograms of the grey value distribution revealed the progression of reaction during the experiment (Fig. 7).
Comparing the histogram of SGC sample with the LGC sample, the shape of the histograms from the LGC sample (Fig. 7a) is clearly asymmetric compared to the bell-shaped histograms in the SGC sample (Fig. 7b).The slope on the precipitation side of the LGC histogram is steep and becomes steeper as the reaction proceeds, indicating dissolution of olivine with little to no precipitation.The slope on the pore side of the LGC histogram becomes shallower due to the secondary porosity generation due to the dissolution of olivine.
Data from the SGC sample showed that the histograms became uniformly flatter and wider over the duration of the experiment, with an increase in the number of both the darkest and brightest voxels at the expense of the voxels with intermediate grey values.These systematic changes in the absorption behavior can only be caused by the sample reacting and indeed reflect the dissolution of olivine, the generation of pore space and the precipitation of reaction products, in addition to phase contrast around newly generated edges in the sample.The best-fit curves to the histograms evolved systematically during the reaction and intersected in a relatively narrow grey value range (−0.000077, −0.000055) (Fig. 7).Voxels with grey values darker than −0.000077 correspond to fluid-filled pores, whose volume proportion increases throughout the reaction process.

Porosity evolution of LGC compared to SGC
Pore space development in LGC is significantly different from SGC.As shown in Figs. 5 and 6, the secondary pore space in the LGC developed simultaneously at the interior and the near-surface area of the sample.However, in the SGC sample, pore space first occurred in the area close to the surface and propagated inwards.
To further examine the difference in porosity generation of the LGC and SGC samples during the reaction, pore spaces within the subvolumes are segmented from the images and calculated.Figure 8 compares the porosity of subvolumes 1 and 2 (Zhu et al., 2016) from SGC and subvolumes 3 and 4 from LGC samples.As shown by the figure, initial porosity of the LGC and SGC samples is similar.The rate of porosity increase in the LGC sample is similar to the SGC during the first 36 h after the reaction started.Porosity of the LGC sample is increasing throughout the duration of the experiment while porosity of the SGC shows a decrease after 90 h of reaction.In the LGC sample, the porosity increased ∼ 5 % and likely resulted from dissolution.This is consistent with other studies that showed the precipitates do not occur in olivine carbonation within 36 h of reaction (e.g., Lisabeth et al., 2017a).
We then further evaluated the porosity distribution within the subvolumes by quantifying the porosity of the 2-D tangential slice along the samples' radial direction.Figure 9 compares the porosity distribution of subvolume 2 from the SGC sample with subvolume 4 from the LGC sample.Subvolume 4 shows a homogeneous increase in porosity of ∼ 5 % throughout the entire subvolume.Subvolume 2, while showing an increase of porosity during the first 30 h of the reaction, starts to show a porosity decrease in areas close to the center of the cup wall after 81 h of reaction.Comparison of the subvolumes from SGC and LGC revealed that the porosity generation within the LGC sample is relatively homogeneous while the porosity generation within the SGC sample starts to show contrast between the cup's near-surface area and its interior after 81 h of reaction.Results from the porosity evaluation further support the finding that the LGC sample is dominated by dissolution with no presence of precipitation.

Fractures in LGC versus SGC
As discussed in Sect.3.1, the planar fractures observed in the LGC sample are likely a result of the dissolution-assisted extensional crack.This is different from the stretching-induced fractures in SGC (see Zhu et al., 2016) where non-uniform precipitation is indispensable.The dissolution-assisted fracturing in LGC and the stretching-induced fracturing in SGC both generated secondary pore space during the experiments, but the fractures in LGC differ distinctively from those in SGC in many ways: -Firstly, the dissolution-assisted fracturing is a single, planar feature in 3-D with no obvious secondary fractures branching out, while the fractures observed in the SGC sample formed a network of intersecting cracks.
Figure 10a shows the morphology of the dissolution plane and the fracture network in 3-D.It is shown in Fig. 10b that the fractures intersected with each other and formed a complex wedge shape network with the vertex pointing towards the sample's interior.The dissolution-assisted fracturing mainly developed along a plane and shows less intersection with other features.
-Secondly, the LGC sample dominated by dissolution features shows clear evidence for shrinkage of larger grains and disappearance of smaller grains at the interior of the aggregate (Fig. 5a).The SGC sample exhibits patches that develop during the reaction which are evidence for the reaction product precipitation (Zhu et al., 2016).
-Thirdly, the development of micron-scale dissolution is simultaneous, with the shrinkage of grains occurring both at the surface and the interior of the cup wall (Figs. 5 and 6a).In contrast, development of the reaction-induced fracturing is successive with most fractures occurring first at the surface and migrating towards the interior of the cup wall.This caused the observed wedge shape of the fracture network (Fig. 10b).Development of the fractures also exhibits a hierarchical sequence with main fractures appearing first.The secondary fractures branch out from the main fracture and divide the sample into smaller domains (Fig. 2 from Zhu et al., 2016).
-In addition, no precipitates were observed along planar dissolution features in the LGC sample; the cup wall remains straight throughout the experiment.But for the SGC, the cup wall shows clear spalling, which is a sign of non-uniform expansion that links to precipitation in this experiment.

Dissolution and fracturing of individual olivine grain
The reaction affected not only the aggregate but also the individual olivine grains.Both the dissolution and fracturing are observed at grain scale in the SGC sample.Figure 11 shows a series of image slices through a nanotomography dataset, moving through an olivine grain in the cup wall.The grains clearly exhibit channels (etch pits) in the reaction zone.In video S1 (see the Supplement), it can be seen that these channels penetrate into and even through olivine grains.As we do not observe any precipitation filling or clogging the tubes, it is therefore reasonable to think that the channels start as etch pits and grow deeper with the advancing of dissolution and eventually become the through-going channels observed here.The tubular shape and the depth of penetration indicate that they are "wormhole" features, likely resulting from dissolution.The shape and the width of these channels vary, with wider inner channel diameters below the surface suggesting more extensive dissolution at depth.On average, most typical tubes have an inner diameter of around 720 nm.
The fracturing is also observed on individual grains at nanometer scale.Figure 12 shows nanotomographic evidence for hierarchical fracturing within olivine grains.The secondary fractures developed from the primary fracture and formed in a direction that is perpendicular to the earlier ones.
In the reconstructed microtomographic images of the loose grains from the SGC, we observed the precipitation of secondary minerals on the surface of the olivine grains inside the cup (Fig. 13).On the basis of their rhombohedral shapes, we identified these as magnesite crystals.However, the phase contrast between the precipitates and olivine grains is very small, and at the current spatial resolution of ∼ 2 µm, we could not segment precipitates from olivine grain with acceptable uncertainties.Other minerals (serpentine, brucite, etc.; see Reactions (R1)-( R4)) were likely also present in the sample but could not be isolated at the given image resolution.Measurement of the magnesite circumference revealed continuous growth during the experiment (see Fig. 5 in Zhu et al., 2016).The first magnesite crystals emerged after 48 h, and grains kept nucleating and growing after that.Growth continued until the experiment was aborted and no deceleration could be observed at any point, which indicates that the sample continued reacting.We determined a growth rate for the grain perimeter of 0.772 µm h −1 , which, by assuming cubic shape of the crystal, yields an equivalent growth rate of 7.18 × 10 −3 µm 3 h −1 .We used a density of 3.01 g cm −3 and molar weight of 84.314 g mol −1 for magnesite (MgCO 3 ) in our calculation.Assuming a specific reaction surface of 50 × 50 µm 2 , this gave a magnesite growth rate of 2.85 × 10 −15 mol cm −2 s −1 .This is in general in agreement with the calculation of Saldi et al. (2009).
The volume change of a loose olivine grain inside the cup was calculated to quantify the competing effect of dissolution and precipitation during olivine carbonation.Individual  grains were labeled from the segmented data and their volumes determined.Figure 13 plots the volume change of a single grain selected from the loose grains inside the SGC over 13 successive microtomographic datasets (covering 146 h).Magnesite overgrowth causes a significant roughening of the olivine grain surface.While the volume of the individual magnesite grain steadily increased throughout the reaction (Zhu et al., 2016), the total volume of the grain (olivine plus precipitates) fluctuates from time to time, which reflects variable rates both in the precipitation of magnesite as well as in the dissolution of olivine.A large drop in grain volume occurred at around 38 h, which is consistent with a period of vigorous olivine dissolution.The largest continuous grain volume increase took place between 40 and 70 h, caused by the precipitation of magnesites.At ∼ 70 h, the grain volume again decreased considerably, indicating that dissolution became dominant once more.This second dissolution episode coincides with the appearance of reaction-induced fractures in the aggregate wall at ∼ 68 h, suggesting a positive feedback process.

Discussion
We claim that our experimental observations indicate the activity of two different mechanisms that both create fluid pathways effectively.These are dissolution-dominated fluid pathway generation at micrometer scale in the case of the LGC experiment and at nanometer scale in the case of SGC, and reaction-induced fracturing in the case of the SGC.We detail our interpretation in the following sections.-d).The vertical distance between each 2-D cross section is ∼ 600 nm.Reacting fluid causes a preferential dissolution of the grain which develops small channels that dig into the grain.These features provide a fluid path and eventually break grains, exposing new reactive surfaces.
Figure 12.Network of microcracks (red) in the reaction olivine cup wall from SGC shows a polygonal pattern.The dissolution channels make the grain (light grey) more susceptible and a stress concentration is also likely to occur around the etch pits.

Dissolution and etch pitting
Dissolution and etch pitting are important mechanisms that affect the grains' surface morphology and the permeability of the sample (e.g., King et al., 2010;Røyne and Jamtveit, 2015).The nanotomographic observation, that etch pitting incurs extensive dissolution beneath grain surfaces, was also documented by Lisabeth et al. (2017a, b) during the carbonation of dunites.Peuble et al. (2018) also observed nanometerscale veinlets forming oblique to subvertical channels in partially carbonated olivine grains during percolation experi-ment.The hollowing out of olivines seems especially important in areas where the grain boundary porosity is decreasing due to the precipitation of secondary minerals.There, subsurface dissolution channels in olivine grains preserve important fluid pathways and maintain the reaction.This supports the hypothesis of Andreani et al. (2009) that the permeability can be maintained by the preferential dissolution even in the cases where the overall porosity is decreasing.Apart from providing access for fluids, subsurface dissolution features also make the grain more susceptible to fracturing and thereby promote the generation of fracturing observed in the experiment.
The permeability of these dissolution channels was calculated using Hagen-Poiseuille's law (Poiseuille, 1844) and Darcy's law (Darcy, 1856).Hagen-Poiseuille's law allows the calculation of the fluid flow inside a tube in dependency of the pressure gradient: where Q is the volume flow/rate of discharge, r is the radius of the tube, η is the viscosity of the fluid, and P is the pressure difference over the flow length L. Darcy's law stated that the rate of discharge is proportional to the viscosity of the fluid and the pressure drop over a given distance, i.e., for a tube with a radius in r: www.solid-earth.net/9/879/2018/Solid Earth, 9, 879-896, 2018 where k is the permeability of the tube.Combining the two equations, k can then be derived as In this experiment, r is ∼ 6 pixels (360 nm).This yields a permeability of ∼ 1.6 × 10 −14 m 2 , which is high enough to transmit reacting fluids through these channels.Considering the density of wormhole features over the grain, the permeability contributed by the channels could be reduced by 2 orders of magnitude but is still able to transmit fluid.These intragranular channels contrast the transgranular fractures.But the development of these etch-pitting dissolution channels also provides a fluid path for the reaction and allows a more extensive degree of alteration of the grain.They weaken the grains, make them more susceptible to disintegration and provide nucleation sites of the later fracturing.While sparse in our data, we think that over geological timescales the contribution by dissolution channels to bulk permeability and the advance of the reaction would be significant.However, on the timescale of our experiments, these features alone are insufficient to explain the observed selfsustainability of the reaction considering the scale and density of the dissolution channels, and we argue that the main contribution must come from volume mismatch cracking in our laboratory study.

Reaction-induced fracturing
Olivine carbonation could produce up to a 44 % increase in solid molar volume assuming the reaction can proceed to completion.If such a volume increase takes place, the crystallization pressure generated could be high enough to fracture the host rock (Kelemen et al., 2013;Kelemen and Hirth, 2012).However, experimental studies on olivine carbonation show no evidence of high crystallization forces (van Noort et al., 2017) but rather suggest that precipitation causes the pore space to fill up and halt the reaction before the crystallizationinduced pressure reaches the critical value needed to generate fracture (Hövelmann et al., 2012;van Noort et al., 2017).Our quantitative estimates indicate that in these experiments, the maximum volume expansion is ∼ 5 %.This is not enough to break the host rock, as shown in salt crystallization experiments (Scherer, 2004).Indeed, the nanotomography data show only dissolution features such as etch pits and wormholes, with no evidence of cracks in olivine grains surrounded by precipitates (Fig. 11).The lack of evidence for crystallization pressure-induced cracking is consistent with other experimental studies (e.g., Hövelmann et al., 2012).
Based on a microtomography experiment, Zhu et al. (2016) proposed an alternative fracturing mechanism during olivine carbonation where crystallization pressure is not high enough to directly break the host rock.In their experiment, the observed fracture patterns in SGC are analogous to desiccation cracks (e.g., Edelman, 1973;Plummer and Gostin, 1981).In Zhu et al. (2016)'s model, the loose olivine grains inside the cup act as precipitate traps that keep the surface of the cup wall relatively free of precipitation.In the interior of the cup wall though, away from the precipitate traps, the crystallization pressure builds up and causes expansion.While the crystallization pressure is too low to cause shear fracturing of the cup, the expanding cup wall interior stretches the surface of the cup wall and causes it to fail in tension and tear.This is facilitated by the near-constant surface area (which decreases slightly as a result of dissolution).In analogy with desiccation cracks, the resulting fractures form characteristic and systematic polygonal patterns.The first set of fractures intersects at right angles, and all subsequent fractures divide the sample into smaller polygonal domains with increasing intersection angles.Since the fracture pattern develops successively rather than simultaneously, the higher-order fractures form in a different stress geometry and as a result migrate perpendicular to the surfaces generated by the previous fracturing event.
To evaluate the potential of surface stretching as a fracture generating mechanism, we estimated the stress that could be produced by the volume mismatch in the cup wall of SGC.We did so by identifying and tracking grains whose spatial coordinates (x, y, z) changed continually as the sample expanded.Distances between these grains were measured at different times.Our measurements indicate an average expansion of ∼ 9.1 µm over a distance of ∼ 260 µm from 7 to 125.9 h after the start of the experiment.The results revealed an axial elongation of 2.78-4.71% in ∼ 120 h in the LGC sample.However, little to no expansion was observed at the near-surface area of the sample's cup wall.This would translate into an axial strain of ∼ 0.03 of the outer layer in order to compensate the volume mismatch.The elastic moduli Figure 14.Pore distribution within the subvolumes from SGC sample.For subvolume 1, which contains a more inner part of the sample, porosity is relatively homogenous among the volume.For subvolume 2, as it contains a more outer part of the sample, a concentration of high porosity can be found in the outer edge compared to the inner edge of the subvolume.This contrast in porosity also reflects a non-uniform precipitation which generates stress that fractures the rock.and strengths of the synthesized porous olivine aggregates are similar to weak sandstones.Using Young's modulus of ∼ 10 GPa yields extensional stresses generated due to the expansion of ∼ 300 MPa, easily exceeding the tensile strength of the sample (∼ 10 MPa).Interestingly, our estimate of extensional stress generated by the volume mismatch is of the same magnitude to the stress from crystallization pressurization (e.g., Kelemen et al., 2013).For natural peridotite, Young's moduli range from 108 to 194 GPa (Christensen, 1966), tensile strength is 50 to 290 MPa, and spall strength is ∼ 58 MPa (He et al., 1996).An equivalent volume expansion of ∼ 10 % in nature could lead to a stress of 3.24 GPa.In both cases, the stress is more than sufficient to fracture the material.However, these are simple estimations of stress and strain made with basic assumptions and local conditions.Considering the extent of the local carbonation reaction and how the expansion in the center is affecting straining of the outer layers, the estimated stress can be considerably lower but should still be in a range that is sufficient to break the material.
To generate the expansion cracks via surface stretching, the volume mismatch must be substantial, which requires keeping the near-surface region free of precipitates.Zhu et al. (2016) suggested that the loose olivine grains inside the sample cup worked as precipitate traps/attractors in this experiment.Because the rate of crystal growth decreases drastically as the curvature of the substrate increases (García et al., 2013;Ziese et al., 2013), large grains in general are preferred sites for precipitation of new crystals.With a size contrast of ∼ 2 orders of magnitude, the loose olivine grains (100-500 µm) in the immediate vicinity of the inner cup surface fulfilled the function of precipitate traps and thereby kept precipitation level at the surface of the olivine cup wall low.
We tested the idea of volume mismatch cracking by conducting the LGC experiment.Now, the size contrast between the grains forming the aggregate in the cup wall and the www.solid-earth.net/9/879/2018/Solid Earth, 9, 879-896, 2018 loose grains inside the cup was significantly reduced, and we expected less efficient precipitate trapping and consequently little to no reaction-induced cracks.The experimental results support this idea.The only planar features observed in the new microtomography experiment are planardissolution-assisted cracking (Figs. 6 and 10a).
A detailed examination of the dissolution channel shown in Fig. 6 revealed no evidence of precipitates there.This places doubt on the crystallization pressure being responsible for fracturing during olivine carbonation.If not, what could be an alternative explanation for the observed fracturing?To further address this question, we examined the porosity evolution and distribution in the 3-D tomographic datasets.
Despite the histogram analysis revealing a bulk increase in the porosity of the cup wall during the experiment, the distribution of these newly generated pores was inhomogeneous in the sample (Figs. 9 and 14).This perturbed the initially homogeneous porosity distribution.This effect became particularly apparent in subvolume 2, which started to exhibit higher porosity where it was closest to the outer surface of the sample after 68 h (see Videos S2, S3 and S4 for details).The difference in porosity distribution within the sample became more pronounced as the reaction proceeded, with porosity in the outer surface of the cup wall increasing while the porosity inside the cup decreased.In our interpretation, this change in the pore volume reflects a contrast in the precipitation rate where precipitation proceeded more slowly in the outer part compared to the inner part of the sample wall, leading to different rates of expansion and the generation of tensile stresses.
The "expansion cracks via stretching" mechanism can explain the observed microstructure evolution in two subvolumes of SGC (Fig. 4).Since subvolume 2 is located at the periphery of the cup, it would be fractured before subvolume 1, which is located in the center of the cup wall.This prediction from the expansion cracks via stretching mechanism is consistent with the observed distribution of pore space that most porous area is located close to the periphery.This mechanism also explains why the fractures tend to develop perpendicular to the reaction surfaces.Similar models relating the reactiongenerated stress and fractures have also been used to explain other exfoliation cracks (e.g., Blackwelder, 1925).
In summary, two fundamental observations from our experiments are inconsistent with the "crystallization pressureinduced fracturing" model and form strong arguments against high crystallization forces (i.e., sufficient to break the surrounding rock) (Figs. 11,14).Firstly, if the cracks were generated directly by crystallization pressure, we should expect them to initiate in a region with intense precipitation and porosity reduction.However, in the outer layer of our sample where the fractures are observed, no precipitates formed prior to the fracturing.Secondly, on the nanoscale, the channels formed by dissolution show hollow and smooth inner surfaces and no precipitation of magnesite (or any other minerals) which shows evidence of low crystallization pressure.
Our detailed analyses provide quantitative support to the "surface cracking via volume mismatch" model first proposed by Zhu et al. (2016).Previous experimental studies on olivine carbonation show that the crystallization force is low (van Noort et al., 2017), suggesting that breaking host rocks by crystallization pressure as in salt crystallization is unlikely, which is in contrast to fracture networks that are commonly observed in naturally occurred serpentinized and carbonated peridotites (Iyer et al., 2008;Macdonald and Fyfe, 1985).We present a resolution to this conundrum by documenting a process that allows fracturing without a high crystallization force.

Coupled mechanisms of dissolution and precipitation-driven fracturing
The findings of this study can be summarized in a mechanism that couples dissolution and precipitation during olivine carbonation.If dissolution and precipitation are heterogeneously distributed in a rock, non-uniform volume expansion can cause breaking of the host rock via surface stretching.In nature, heterogeneity in the porosity and permeability of a rock formation is common, which may cause nonuniform concentration of reaction and distribution of precipitation (Wells et al., 2017).As shown in our study, the resulting volume mismatch could lead to expansion fractures.The fractures provide new fluid pathways and expose fresh reactive surfaces to sustain the carbonation.In the long term, fluid pathways may be provided by wormhole etch pitting.Dissolution channels could deteriorate rock strength over longer timescales (Fig. 15).In a field study on serpentinization of peridotites at ocean ridges, Rouméjon and Cannat (2014) show that the reactioninduced fracturing occurs in the early stages of serpentinization (probably before 20 % of serpentinization) while the rest of the volume increase is accommodated by the serpentine itself and dissolution processes dominate until completion of the reaction.Clearly, natural processes are more complex.Laboratory studies are not capable of simulating nature settings due to the vastly different length and timescale.Instead, we focus on understanding the underlying mechanism of porosity generation during olivine carbonation reaction.Using the coupled dissolution and fracturing model described above, we postulate that at the beginning of a carbonation reaction, dissolution plays an important role in maintaining porosity.As reaction progresses, non-uniform expansion of the rock due to precipitation could lead to fractures.Once the host rock is fractured, accelerated reaction takes place to achieve 100 % carbonation.
In general, several different mechanisms seem to facilitate olivine carbonation and contribute to sustaining it.On relatively short timescales, rapid reaction-induced tensile fracturing could be the dominating mechanism that maintains the reaction, whereas on a longer timescale, dissolution and the formation of channel-like structures may dominate.

Conclusions
Using synchrotron-based micro-and nanotomography, we documented and quantified the reaction progress during olivine-fluid interaction on the micron scale.This allowed us to identify mechanisms that sustain the reaction despite its large positive volume change.
In summary, our experiment results suggest the following: -The reaction-induced fracturing observed in our experiments results from non-uniform volume expansion.Tensile stresses arise from heterogeneous precipitation and the resulting contrast in the expansion between a faster-expanding interior and a slower expansion near the surface.
-Even though the dissolution cannot be used alone to explain the sustainability of the experimental timescale olivine carbonation, it provides evidence that dissolution etch pits can provide fluid path and fresh reaction surface for the reaction to proceed.This helps in explaining the naturally occurring complete alteration of peridotite, as the timescale for natural carbonation ranges from thousands to millions of years.Even if the dissolution channelizing would only allow slow fluid flow, it could still induce significant alteration given time.
-The coupled mechanism of dissolution and reactioninduced fracturing accounts for maintaining the reaction processes during olivine carbonation.It explains on different time and space scales the formation of observed natural outcrops of completely carbonated peridotite.
-The results from our experimental study also provide new insights into the application of CO 2 mineral sequestration.

Figure 1 .
Figure 1.Illustration of the mechanisms of reaction-induced fracturing during olivine-fluid interaction.(a) The crystallization pressurization model describes the development of fractures caused by crystallization forces exerted on the surroundings due to growth of precipitates.Salt crystallization (Scherer, 2004) is a typical example of the crystallization pressure-induced fracturing.The fractures first appear at areas where precipitation is most concentrated and propagate outwards.(b) The surface cracking model describes the development of fractures as a result of a contrast in expansion which causes stretching at the surface.A difference in the precipitation rate between the periphery and interior of the sample causes them to expand at different rates with the interior expanding faster than the surface.This builds up the tensile stress at the surface that fractures the sample and leads to a development of a polygonal fracture network.The fractures propagate from the surface inwards.

Figure 2 .
Figure 2. Illustration of sample configuration.(a) The sample is composed a sintered olivine aggregate cup with loose olivine sand fillings (grain size 100-500 µm).Quantitative microtomographic analyses are conducted on the tangential (pink), radial (orange) and cross (blue) section of the sample.(b) Example of the tomographic image that was used in the 2-D examination of Zhu et al. (2016).

Figure 3 .
Figure 3. Experimental setup for dynamic microtomography.Inside the X-ray transparent pressure cell, the confining pressure, pore pressure and temperature can be controlled independently.The synchrotron radiation imaging records radiographs of the sample at in situ conditions with ongoing reaction at different angular positions with the sample being rotated.

Figure 4 .
Figure 4. Positions of the subvolumes in the cup wall of (a) SGC and (b) LGC, where quantitative 3-D microstructure analyses were performed.In the SGC sample, subvolume 1 (bottom box) is located at the center of the cup wall.Subvolume 2 (top box) is located adjacent to the outer rim of the wall.In the LGC sample, both subvolume 3 (bottom box) and subvolume 4 (top box) are located at the center of the cup wall.

Figure 5 .
Figure 5. Reconstructed images showing the cross-section view of the sample undergoing olivine carbonation.(a) In the LGC sample, original olivine grains shrink, and secondary pore space appears (yellow arrows), suggesting dissolution of olivine.Most of the large pore spaces concentrate at the interior of the cup wall (highlighted by yellow dashed lines).(b) In the SGC sample, the fractures first developed at the surface of the cup and propagated from the outer rim into the cup wall (highlighted by yellow dashed lines).Larger pore space distributes mainly near the rim.

Figure 6 .
Figure 6.(a) Dissolution features in the cross-sectional images of the LGC sample.3-D examination reveals that these dissolution features (blue arrowheads) are associated with planar fractures (yellow arrowheads) appeared in (b) the radial and (c) the tangential section images.The boarder color of each section corresponds to color code given in Fig. 2.

Figure 7 .
Figure 7. Best fits for the grey value distribution histograms of the sample at different stages of the reaction in (a) subvolume 4 from LGC sample and (b) subvolume 2 from SGC sample.Different colors represent time lapses as shown.Pores, olivine and precipitates are identified based on their grey value ranges.Higher values correspond to lighter grey (solids).The more negative a value is, the darker the grey color becomes (e.g., pores are black).Histograms of the SGC sample were calculated from a 2-D image(Zhu et al., 2016), whereas the histograms of the LGC sample were calculated from 3-D datasets.

Figure 8 .
Figure 8.Comparison of porosity evolution in LGC (this study) versus SGC reported by Zhu et al. (2016).
Figure 9. Porosity profiles of (a) LGC and (b) SGC along the radial direction of the sample cup.Different colors indicate reaction time lapse.

Figure 10 .
Figure 10.The 3-D geometry (in red) of (a) dissolution-assisted fracture in LGC is distinctively different from (b) the stretching-induced fractures in SGC.Both displayed volumes are 260 × 260 × 260 µm 3 in size.

Figure 11 .
Figure 11.Reconstructed images from the nanotomography data of the SGC sample's cup wall demonstrate the existence of etch pits and dissolution channels (white arrows) formed in the olivine grain.The precipitates (darker grey) partially fill the pore space (black) between olivine grains (lighter grey).The yellow dashed line (a) marks the orientation of the cross sections (b-d).The vertical distance between each 2-D cross section is ∼ 600 nm.Reacting fluid causes a preferential dissolution of the grain which develops small channels that dig into the grain.These features provide a fluid path and eventually break grains, exposing new reactive surfaces.

Figure 13 .
Figure13.Volume change of an individual olivine grain (total volume of olivine and precipitates) during carbonation reaction in the SGC experiment.The precipitates are idiomorphic and referred to as magnesites (seeZhu et al., 2016).This observed volume change results from a combination of the dissolution of olivine and precipitation of magnesites.Fluctuations of the grain volume manifest the altering dominance of dissolution versus precipitation.

Figure 15 .
Figure 15.Illustration of porosity generation mechanisms during the olivine carbonation reactions.A combined mechanism of surface cracking and the dissolution channelization plays an important role in the porosity generation.Heterogeneity in the microstructure of the material would cause non-uniform distribution of precipitation.This would lead to the generation of surface cracking via volume mismatch and generate secondary porosity.Dissolution also produces pore spaces and fluid pathways through etch-pitting channelization which makes the grains susceptible to the cracking on a longer timescale.