Chemo-mechanical behavior of clay-rich fault gouges affected by CO 2 -brine-rock interactions

: The impact of long-term CO 2 -brine-rock interactions on the frictional properties of faults is one of the main concerns when ensuring safe geological CO 2 storage. Mineralogical changes may alter the frictional strength and seismogenic potential of pre-existing faults bounding a storage complex. However, most of these reactions are too slow to be reproduced on laboratory timescales and can only be assessed using geochemical modeling. We combined modeling of CO 2 -charged formation water and fault gouges (1–1000 years residence time, i.e. 10–10 6 pore volume flushes) with friction experiments on simulated fault gouges ( T = 22–150 ° C; σ neff = 50 MPa; P f = 25 MPa; V = 0.2-100 μ m/s), having mineralogical compositions as predicted by the models. As an analogue for clay-rich caprocks overlying potential CO 2 storage sites in Europe, we used the Opalinus claystone. Our experiments showed that, although significant mineralogical changes occurred, they did not significantly change the frictional behavior of faults. Instead, initial fault-gouge mineralogy imposed a stronger control on clay-rich fault behavior than the extent of CO 2 -brine-rock interactions, even under chemical conditions allowing for significant reaction. We demonstrated that the impact of mineralogical changes due to CO 2 -brine-rock interactions on the frictional behavior and seismogenic potential of faults could be assessed using our combination of geochemical modeling and friction experiments. Note that a complete understanding requires evaluation of additional effects, such as that of shear velocity, effective normal stress, and other fault characteristics (maturity, shear strain). C (cid:2) 2018 The Authors. Greenhouse Gases: Science and Technology published by Society of Chemical Industry and John Wiley & Sons, Ltd. Additional supporting information may be found online in the Supporting Information section at the end of the article. behavior in the subsurface


Introduction
O ne of the most promising technologies to reduce global CO 2 emissions is geological storage of CO 2 , in addition to transitioning to renewable energy sources, improving energy efficiency, and reducing energy consumption. Carbon capture and storage (CCS) entails the capture of CO 2 at source, and transport to, and injection into, a suitable storage site. The main storage mechanism is structural trapping of the liquid or supercritical CO 2 phase directly underneath the caprock. 1 Over time, additional, but generally more limited, storage space can be provided by dissolution trapping of CO 2 into the formation brine, 2 residual trapping of CO 2 within the pore space 3,4 and/or mineral trapping if the host mineralogy is favorable. 5 In recent years a significant amount of research has been conducted to understand CO 2 -brinerock interactions that take place during long-term CO 2 storage in saline formations and depleted hydrocarbon reservoirs (see Kampman et al. 6 for a review). However, many storage sites contain, or are bounded by, faults. It has been recognized that CO 2 -charged formation water can react chemically with fault minerals, thereby potentially altering fault friction behavior and creating a seismicity hazard. 7 The risk of induced seismicity due to CO 2 storage should therefore be evaluated prior to geological storage of CO 2 to mitigate anthropogenic greenhouse gas emissions. 8 Pre-existing faults that penetrate the caprock may be reactivated during CO 2 injection and lead to seismicity if fault slip is unstable. 8,9 In the short term (i.e. during the injection period), fault slip can be induced if CO 2 injection reduces the normal stress and/or increases the shear stress acting on the fault plane. Stress changes may be caused by pore pressure changes in the fault, by the coupled poro-elastic response of the surroundings (i.e. reservoir and caprock), and by thermally induced local stress changes. 10,11 Carbon dioxide injection and storage may also induce other chemo-mechanical processes that impact fault gouges and surrounding host rock, such as changes in rock strength and fracture toughness induced by wet supercritical CO 2 , 12-14 co-injection of acid gases (e.g. H 2 S and SO 2 ), and/or development of geochemical domains on mineral surfaces (e.g. water films) and in nano-scale pores. However, our study develops a foundational understanding of mineral behavior due to CO 2 -brine-rock interaction on which detailed future studies can build.
In clay-rich caprocks that overlie proposed storage reservoirs, such interactions can involve dissolution and re-precipitation of carbonates, dissolution of plagioclase feldspar, and clay minerals (illite, chlorite, and kaolinite) and precipitation of smectite. 15,16 Mineralogical changes such as these might alter the frictional behavior of fault gouge, i.e. the wear product created by fracturing of intact host rock. A change in the frictional strength affects the ease with which a fault is reactivated, while a change in the slip stability has implications for induced seismicity. The importance of the mineralogy for the frictional behavior of fault gouge has been demonstrated by numerous experimental studies (e.g. see Ikari et al.). 17 Competent minerals (e.g. quartz, feldspar, carbonate and anhydrite) are known to exhibit a high frictional strength (friction coefficient, μ = 0.6-0.85) and both stable and unstable behavior, whereas phyllosilicates (clay minerals and micas) are significantly weaker (μ = 0.2-0.5) and predominantly frictionally stable ( Fig. 1A-B [18][19][20][21][22][23][24][25][26][27][28][29][30][31]. Faults generally contain polymineralic mixtures of these competent and weak minerals, and therefore exhibit complex frictional behavior, where the strength is typically controlled by the dominant mineral(s), 32 as can also be seen in Fig. 1.
To date, a limited number of experimental studies have investigated the effects of CO 2 -brine-rock interactions on the mechanical properties of reservoir rocks, caprocks, and faults, including both short-term experiments (lasting hours or days -see, e.g., Samuelson and Spiers, 33 Pluymakers et al. 34 ) and studies investigating the effect of long-term CO 2 exposure (i.e. post abandonment of the field). 12,13,[35][36][37] One of the challenges encountered in laboratory studies investigating the long-term effects of CO 2 exposure is the slow reaction of aluminosilicates and the storage integrity timescales required by many regulators (>10 000 years), which are typically much longer than available laboratory timescales (weeks, months or occasionally a few years; see Rochelle et al. 38 and references therein).
In this paper we overcome this problem by using geochemical modeling to predict mineralogical changes within faults at timescales not achievable in the laboratory. Based on previous work, 20,35,39 it is expected that frictionally weak or unstable mineral phases, such as clays or carbonates, would need to make up between 20% and 80%, respectively, of the gouge composition to significantly affect its frictional

Geochemical simulations of CO 2 -brine-rock interactions in clay-rich fault rock
This study focuses on chemical-mechanical behavior induced by mineral dissolution and precipitation due to long-term CO 2 -brine-rock interactions. Using Geochemist's Workbench 11.0 (GWB), the b-dot ion association model, and the resident thermo.tdat database 42 according to previously described methods, 43,44 three distinct conceptual scenarios were developed to understand how fluids rising from an underlying sandstone storage reservoir may impact chemical-mechanical processes within a faulted carbonate-bearing claystone caprock. The first scenario is a baseline scenario in which no CO 2 is injected into the underlying storage reservoir. Reservoir water that invades the fault within the overlying claystone cap is therefore not saturated with CO 2 (RW-scenario series, cf. Tables 1 and C1 of the supplementary material). In the second scenario, CO 2 is injected into the reservoir, and the reservoir water therefore becomes saturated with CO 2 . Once CO 2 -saturated water invades the faulted caprock, the fault is assumed to form a closed system, i.e. the fault is isolated from the underlying reservoir. The third scenario is in essence the same as the second but the fault behaves as an open system with respect to the CO 2 (but not the water) residing in the CO 2 plume present directly below the caprock, i.e. CO 2 will continue to access the fault to keep the pore fluid saturated with CO 2 . Dissolved CO 2 is consumed by water-rock interactions in the closed system, whereas dissolved CO 2 is buffered and CO 2 saturation of water in the fault remains constant in the open system (C-and O-scenario series, respectively; cf. Tables 1 and C1 of the supplementary material). Details of the geochemical model and its input parameters are summarized in Appendix A of the supplementary material. For each of these three reservoir scenarios, we approached fluid replacement within the fault by so-called fault-valve behavior, [45][46][47] where the fluid within the fault is periodically replaced. For the reservoir scenarios, both the fluid residence time in the fault (i.e. the time that one pore volume of reactive fluid resides in the fault rock prior to replenishment with a fresh pore volume, t r ) and total time of fluid-rock interaction within the fault (total reaction time investigated in each scenario, T r ) were varied. We simulated four residence times, 1, 10, 1000, and 10 000 years, to represent both rapid and more gradual overpressurization of the reservoir, and periodic reactivation of the fault assuming typical clay gouge permeabilities (10 −17 -10 −22 m 2 ), 48 and total time of fluid-rock interaction of 10 000 years and 1 million years. The number of pore volumes of reservoir water that passed through the fault for the duration of the simulation time (the ratio of the total time of fluid-rock interaction and the residence time, T r /t r ) ranges from 10 to 10 6 ( Table 1).  Pyrite was not taken into account for this ratio and it was adjusted accordingly. c OPA consisted for 5.2% of illite and 24.4% of illite-smectite clay. CFOPA consisted of 7.0% of illite and 33.0% of illite-smectite clay (see also In our evaluation of the modeling results, we grouped the various minerals in the gouge according to four main mineral groups, depending on hardness and crystal morphology: quartz, carbonates (calcite, dolomite, siderite), micas (chlorite, paragonite), and clay minerals (illite, smectite, kaolinite). Note that, due to limitations of geochemical models and experimental methods, we treated interlayer illite-smectite as a mixture of end-member illite and smectite in appropriate proportions (i.e. 80% illite, 20% smectite, cf. Table A1). We do not expect this approach to affect our geochemical modeling results or the frictional properties of our simulated gouges significantly (cf. Moore and Lockner). 49

Sample preparation
All friction experiments were performed on finely powdered OPA or mineral mixtures. Intact natural materials and individual minerals were crushed using a mortar and pestle, and sieved to a grain size of <35 μm. For selected compositions ( Table 1), powders of quartz (99.9% pure; Beaujean quartz sand), kaolinite (100% pure; Merkx R ), muscovite (90% muscovite plus ß10% quartz impurities for which no correction was made; commercially obtained), calcite (Iceland spar 23 ), dolomite (97% dolomite plus ß3% labradorite; Forno-Apuan Alps), and illite-rich shale (Rochester shale, 65% illite plus 35% quartz impurities 29 ) were mixed by weight and thoroughly stirred to ensure a homogeneous mixture. Siderite, Ca-nontronite, and paragonite were difficult to obtain; analogue mineralsrespectively dolomite, Ca-montmorillonite, and muscovite -were used instead. Pyrite was omitted due to its small proportion in OPA and the mineral proportions were corrected accordingly.
In addition to using mineral mixtures, a carbonate-free OPA (CFOPA) was prepared by exposing crushed OPA to an excess of acetic-acid (C 2 H 4 O 2 ; pH 3, at ß20-80°C), effectively removing any naturally present carbonate. After this dissolution process, the material was washed with demineralized water, dried, and sieved again (<35 μm).
Intact, natural OPA possesses a broader range of grain size than the prepared gouges, with calcite fossils of up to 200 μm in size (see Fig. 2). However, these larger calcite clasts will most likely be crushed during fault gouge formation in nature. 50 The prepared gouges therefore have a size range typically observed in both simulated and natural fault gouges. 51

Experimental apparatus, procedure and data processing
Frictional sliding experiments were performed using a rotary shear apparatus, as previously described by Niemeijer et al. 52 and den Hartog et al. 31 For the experiments, a ß1 mm thick ring-shaped gouge layer (inner radius 11 mm, outer radius 14 mm) was sandwiched between two roughened pistons. Under constant effective normal stress (σ n eff ) and pore fluid pressure (P f -distilled water, fully saturated conditions), the layer could be sheared at a constant rate, while the generated shear stress (τ ) was measured externally. A total of eight friction experiments were performed (σ n eff = 50 MPa, P f = 25 MPa), each successively at room temperature (ß22°C), 100°C and 150°C. The temperature range investigated is relevant for the depth range of potential CO 2 storage reservoirs and includes the transition from stable to unstable frictional behavior at 100-150°C as observed for calcite. 23 Note that due to this temperature-stepping approach, thermal expansion of the apparatus prevented us from measuring the compaction or dilation behavior of the gouge accurately. At each of the three temperatures, sliding was initiated at 5.43 μm/s, followed by a velocity stepping sequence (0.22-1.086-10.86-1.086-3-10-30-100 μm/s) to investigate the velocity-dependence of friction of the gouge material, reaching a total (corrected) displacement of ß33-34 mm. After velocity stepping at 150°C, a normal stress-stepping sequence (50, 60, 80 and 100 MPa for 2 mm per step, sliding velocity of 10 μm/s) was performed to determine gouge cohesion.
Shear displacement (resolution ± 1 μm, respectively), normal force (± 0.05 kN), torque (0.1% of full scale; 1.2 N), pore fluid pressure (± 0.015 MPa), and sample temperature (± 1-2°C) were logged. Torque and normal force were corrected for predetermined seal friction to obtain the shear stress at a radius of 12.5 mm and effective normal stress acting on the gouge layer, whereas the shear displacement was corrected for machine stiffness (distortion). The apparent friction coefficient μ ('friction coefficient' hereafter) was calculated as the ratio of shear stress to effective normal stress, μ = τ /σ n eff , effectively ignoring cohesion. The velocity dependence of μ was interpreted in the framework of rate-and-state dependent friction theory [53][54][55] (RSF) and expressed in terms of (a-b) (for details see Appendix B of the supplementary material).
In essence, if the friction coefficient μ increases following an increase in sliding velocity, the fault will slip stably, i.e. the fault exhibits 'velocity-strengthening' behavior and the accelerating slip needed for seismogenesis cannot occur. This is typically characterized by a positive (a-b)-value. 56 However, if μ decreases with increasing velocity, the fault has the potential to yield unstable behavior, indicated by negative (a-b) values. 56 This is called 'velocity-weakening' behavior and it is expressed during the experiment as possible repetitive stick-slip events or seismogenic slip, provided that the elastic stiffness of the loading system falls in the right range. 53 Errors for (a-b) are generally on the order of ±0.001, but may be as high as ±0.0028, depending on the quality of the fit made to the velocity-stepping data to derive the RSF-parameters (see Niemeijer and Vissers, 57 and Appendix B of the supplementary material).

Microstructural preparation and analysis
The experiments were terminated by halting shearing, cooling down the setup, and removing the pore pressure and normal stress. Upon removal from the ring shear apparatus, the sample assembly was dried at 50°C for at least 24 hours. Retrieval of the sample from the assembly led to the partial disintegration of the ring-shaped gouge layer into platy fragments. The larger fragments were cast into resin. Thick sections were prepared manually, perpendicular to the shear direction, and analyzed using a scanning electron microscope (SEM). To ensure careful polishing of the sections, we used a focused ion beam (FIB-SEM) to ion-mill the section surface, without distorting the sample surface. The SEM analysis was executed in backscatter electron (BSE) mode, to provide the best contrast between the various mineral phases.

Results and discussion
Mineral assemblages for the simulated fault gouge infiltrated by CO 2 -free reservoir water and CO 2 -saturated water are summarized in Table 1 and Appendix C of the supplementary material; the temporal evolution of simulated gouge compositions is provided in Appendix D of the supplementary material. The selection criteria for the mineral compositions tested in the friction experiments are: (i) > 5 wt% difference per mineral group (quartz, carbonates, mica, clay minerals) and/or (ii) > 5 wt% change in competent (quartz + carbonates) versus weak phase (mica + clay) ratio (Table C1). The scenario with the smallest mineralogical change was selected in the case of roughly similar mineralogical changes, representing the minimal impact of mineralogical changes on frictional behavior. Note that the preparation of simulated gouges with compositions similar to natural gouges has been shown to introduce an error of less than 10% in frictional strength, 58 making the approach ideal for assessing the impact of long-term fluid-rock interactions in faults.
As a baseline for evaluating the effect of CO 2 exposure on the frictional behavior of OPA, one scenario of a fault gouge infiltrated with CO 2 -free reservoir water was selected for experimental investigation (cf. Table 2). From the CO 2 -saturated water scenarios, we selected a total of four compositions that showed mineralogical changes significant enough to influence the frictional behavior of the gouge (Table 1). CO 2 -brine-rock reactions significantly enhanced porosity in open-system simulations, whereas in closed-system simulations porosity remained more-or-less constant. However, similar mineral assemblages developed in both sets of simulations for a given residence time and total reaction time (Table C1, supplementary material). As porosity change cannot be accounted for in the type of experiments presented here, only closed-system simulated gouges were tested. We compare modeling and frictional testing results to the measurements on natural OPA to evaluate the effect of reaction with (CO 2 -saturated) reservoir water on mineralogy and how such alterations affect frictional behavior.
The key results of the friction experiments on gouges with selected compositions are listed in Table 2 Table B1 of the supplementary material, and the (a-b) values are visualized in Fig. 4D as a function of composition. In general, friction coefficient versus displacement graphs show an initial increase in μ with shear displacement, followed by either a peak strength and successive drop in μ or by yielding, often with subsequent strain hardening ( Fig. 3A and B). Near steady-state sliding behavior is achieved after ß2-4 mm displacement at each temperature. Strain hardening is suggested to be related to preferential extrusion of relatively weak phyllosilicates from the unconfined sample assembly. 59 The effect of this experimental artefact is still Table 2. List of experiments and key data. μ = (apparent) friction coefficient at displacements of 4.0, 14.7 and 25.4 mm at room temperature, 100 and 150°C, respectively. μ i = internal friction coefficient, S o = cohesion, as determined from linear fits to the data in Fig. 3C and D. For negative (a-b) 4A). However, for completeness, we have also visualized the friction coefficients measured prior to each velocity step (Fig. 4B).
In the calculation of the friction coefficient, gouge cohesion is ignored (see section on 'Experimental apparatus, procedure and data processing'). To determine whether this assumption is justified, shear stress is plotted against the effective normal stress that was applied in the normal stress stepping sequence (Fig. 3C-D). The data can be described by a linear relationship of the form τ = S 0 + μ i . σ n eff , where μ i is the internal coefficient of friction and S 0 is the gouge cohesion. Cohesion was mostly less than 2.2 MPa, though negative values for the cohesion were also obtained (up to −1.8 MPa). This suggests that the error in fitting a straight line to our τ vs. σ n eff data is relatively large, probably due to the extrapolation from > 50 MPa σ n eff down to zero. It is therefore not possible to correct for cohesion accurately. However, cohesion was mostly < 10% of the shear stress for our experiments, except for C-10-1M where it was 4.0 MPa, (ß25-30% of the shear strength), supporting our use of an apparent friction coefficient (see 31,35,57 ).

Effect of interaction with CO 2 -free reservoir water
For the natural OPA, the friction coefficient shows a weak increase with sliding velocity and shear displacement, and a negligible effect of temperature, reaching values of 0.34-0.40 at room temperature and 100°C (average value of 0.37), and 0.34-0.42 at 150°C (average value of 0.38), respectively ( Table 2; Figs 3A and 4A-B). Natural OPA also shows positive (a-b) values, i.e. velocity-strengthening behavior (Fig. 4C-D), typical for wet, carbonate-bearing, clay-rich fault gouges (e.g. Chen et al.). 60 In most scenarios, reactions between CO 2 -free reservoir water and OPA fault gouge begins to dissolve calcite, chlorite, illite, and kaolinite and precipitate mica and smectite (Table 1 and Appendix C in the supplementary material). In two scenarios, frequent MPa, P f = 100 MPa and room temperature, 100 and 150°C. Sliding velocity is varied from 0.22 to 100 μm/s, as indicated at 100°C for RW-10-10k. A: Friction curves for OPA, carbonate-free OPA (CFOPA) and the modeled composition for natural OPA exposed to reservoir water (RW-10-10k). B: Friction curves for selected model scenarios, assuming interactions between natural OPA and CO 2 -saturated reservoir water. Peak or yield values of the shear stress obtained during the normal stress stepping sequence (not shown in A and B) as a function of effective normal stress for C: natural OPA and RW-10-10k, and D: for selected closed systems scenarios reacted with CO 2 -saturated reservoir water.
replacement of reacted pore water with fresh reservoir water (t r = 1-10 years, 10 5 -10 6 pore volumes) completely dissolves both calcite and illite and precipitates kaolinite and a much larger amount of smectite (39-72 wt%; saponite plus nontronite content) within the fault after 1 million years (cf. RW-1-1M and RW-10-1M). As a result, these latter compositions are clay dominated (> 80 wt% clay), which will largely control their frictional behavior, as seen in previous studies (e.g. see Tembe et al. 19 ). Therefore, and given the similarity in composition between the majority of the modeled results (RW-10-10k, RW-1k-10k, RW-1k-1M, RW-10k-10k, and RW-10k-1M), we have only selected RW-10-10k for frictional testing. With μ-values of 0.39, 0.43 and 0.47, at room temperature, 100°C and 150°C, respectively, the OPA-reservoir water scenario RW-10-10k suggests an 11-31% increase in the friction coefficient within 10 000 years, despite a 2 wt% decrease in total frictionally competent phases (quartz + carbonates -see Table 1). The velocity dependence of friction showed a small decrease in (a-b) values for RW-10-10k compared to the natural OPA sample, although values are still positive (Fig. 4C-D).   Table 1 for compositions). C: Average velocity-dependency (a-b) and D: velocity-dependency (a-b) for all velocity up-steps again as a function of temperature for the different compositions tested. From previous discussions: steady-state was not really achieved for some velocity steps.

Effect of interaction with CO 2 -saturated reservoir water
Overall, simulated CO 2 -exposed gouge compositions showed that the μ-values increased with increasing temperature and shear displacement, whereas (a-b) decreased with increasing temperature (cf . Figs 3 and  4), at the same relative displacement compared to unreacted OPA. In terms of mineralogical changes, CO 2 -saturated reservoir water dissolves nearly all carbonates and micas for three scenarios in which fresh fluid frequently replaces reacted pore water within the gouge, t r = 1 year for 10 000 years and 1 million years reaction time (10 4 and 10 6 pore volumes) and t r = 10 years for 1 million years reaction time (10 5 pore volumes) ( Table 1 and Appendix C in the supplementary material). The resulting mineral assemblage is essentially a quartz-clay gouge similar to carbonate-free OPA (CFOPA). The CFOPA showed near-temperature-independent friction coefficients of 0.28-0.30 (Table 2), i.e. 17-20% lower than for natural OPA and 28-36% lower than for reservoir water-reacted OPA (RW-10-10k; cf. Figs 3A and 4A-B). Given that CFOPA contains only 31% quartz as opposed to ß50% quartz plus carbonate for natural and reservoir water-reacted OPA, this behavior is consistent with that typically seen for gouges consisting of a mixture of weak phyllosilicates and competent quartz and/or calcite. Notably, a decrease in μ similar in magnitude to that found here has been reported for an increase in the phyllosilicate phase from ß25% to 50%, as the stress-supporting framework of quartz and/or carbonate clast changes to a phyllosilicate-dominated matrix embedded with disperse clasts. [19][20][21] Values of (a-b) for CFOPA are positive and within the range measured for the reservoir water-reacted OPA and only slightly lower than those for natural OPA (Fig. 4C-D). For longer residence times (fewer pore volumes), CO 2 -saturated reservoir water dissolves micas and increases quartz and clay content within the OPA fault gouge, although the detailed mineralogical evolution of the simulated gouges is complex (Table 1 and Appendixes C and D of the supplementary material). Overall, for t r = 10 years (10 3 and 10 5 pore volumes; C-10-10k, C-10-1M, respectively) the total weak phase decreases by up to 3%, whereas for t r = 1000 years (10 and 10 3 pore volumes; C-1k-10k and C-1k-1M, respectively) it decreases by ß10%. All four simulated gouge compositions showed μ-values increasing with temperature and shear displacement (cf. Fig. 3B and 4A-B), reaching values in the range 0.40-0.54, i.e. 14-50% higher than OPA and up to 21% higher than reservoir water-reacted OPA, when comparing values reached at the same temperatures in these experiments. Regarding the velocity dependence, a weak trend is seen from generally positive (a-b) values at room temperature, towards near zero (a-b) values at 100°C and finally slightly negative (a-b) values at 150°C, particularly at low sliding velocities (V ࣘ 10 μm/s), suggesting a transition towards more unstable type behavior with increasing temperature (see also Table B1 and Fig. B2 in the supplementary material). Note, that the observed repeatability of μ for composition C-10-10k is within ß0.03, which agrees with the reproducibility seen previously for this machine (unpublished data, SAM den Hartog -cf. Fig. 4B, D).

The relationship between frictional behavior and microstructure
Scanning electron microscopy images (backscatter electron mode) of samples C-1k-10k and C-1k-1M, representing end-member samples in terms of their quartz: carbonate ratio (26:32 and 35:22 respectivelysee Table 1) show different internal deformation features (Fig. 5). Sample C-1k-10k (Fig. 5A-B) appears to be homogeneously deformed, with intact quartz grains randomly distributed throughout the gouge layer. On the other hand, hardly any intact calcite grains, which are the predominant and clearly observable phase of carbonate, remain within the body of the gouge layer. Finely crushed calcite is observed (Fig. 5B), particularly surrounding larger quartz grains, aligned along the R 1 -direction. Throughout the body of the sheared gouge, large, fractured, but not crushed calcite grains are observed. This grain-fracturing is absent in the starting material (see the relatively undeformed portions of the gouge at the top and base of the image). Overall, it can be seen that, although extensive fracturing of the calcite grains is obvious, it has not led to the complete pulverization of the carbonate phase. This microstructure is in contrast to the more quartz-rich gouge (C-1k-1M; Fig. 5C-D), which still contains large quartz grains. However, the large quartz grains are embedded in a very fine-grained matrix of crushed carbonate, fragmented quartz grains, and clay particles, and no recognizable intact carbonate grains can be observed.
As stated above, the observed frictional behavior of fault gouges is assumed to depend on the fraction of frictionally weak phases present in a gouge (Table 1 and Fig. 4; see also Tembe et al.). 19 This typical behavior of polymineralic gouges roughly explains the observations for OPA and CFOPA but it fails to account for smaller differences in the friction coefficient such as those seen for C-10-10ka/b and C-1k-1M versus RW-10-10k and C-1k-10k (Fig. 5), the former showing systematically higher friction coefficients. It should be noted that frictionally stronger samples are characterized by higher quartz: carbonate ratios. Based on the microstructural observations, we speculate that the frictional behavior may also depend on the proportion of quartz versus carbonates in the competent phase, which influences the frictional behavior of a gouge during shearing. Our microstructural evidence suggests that the gouges with proportionally higher quartz content will generate a larger fraction of finer grained carbonate material. This is most likely caused by the preferential fracturing of calcite grains along the main cleavage plane (c-axis) over that of conchoidal fracturing of quartz. 61,62 This cleavage weakens the calcite grains, leading to an overall reduction in the carbonate grain size compared to quartz, impacting the frictional behavior, potentially aided by dissolution-induced grain-size reduction. The smaller size of the carbonate grains promotes thermally activated compaction behavior through dislocation glide and/or diffusion, resulting in a lower frictional resistance and potentially velocity-weakening behavior as observed in these samples at higher temperatures. This velocity-weakening behavior is in line with the behavior seen for pure calcite at 100-150°C. 23 Figure 5. Scanning electron micrographs (BSE-mode) illustrating the microstructure of sheared, simulated fault gouges. A: The sheared C-1k-10k sample, consisting of 20% calcite and 12% of dolomite and siderite grains, shows crushed calcite (± siderite; white) next to intact and larger quartz grains, as well as fractured but not crushed calcite grains, all throughout the body of the gouge layer. Crushed calcite appears to form R1-shear bands. Note that the boundary shear zones are not fully developed. B: Inset of (A), showing part of the main body with distributed intact quartz next to internally fractured and crushed calcite grains. C: Sheared C-1k-1M sample, consisting of 6% of calcite and 16% of dolomite and siderite, showing some intact coarse quartz grains embedded in a matrix of clay and crushed calcite (white) and crushed quartz grains. Calcite appears to align along the R1-shear band orientation. Note the absence of a very distinct boundary shear zone. D: Inset of (C), showing significantly reduced calcite next to large quartz grains.

Impact of CO 2 -brine-rock interactions on frictional behavior of clay-rich faults
The extent of reaction in CO 2 storage systems is controlled on one hand by the initial (i.e. pre-CO 2 -exposure) mineralogy 58 and on the other hand by the fluid: rock ratio, i.e. the rock porosity and fluid flux through it. 36,63 For a range of fluxes of CO 2 -saturated brine, geochemical modeling of the porous Opalinus claystone fault gouge showed a complex mineralogical evolution, including dissolution of carbonate minerals and precipitation of phyllosilicates and quartz, not seen in the presence of reservoir brine alone (see Fig. 6A). All four CO 2 -reacted mixtures show that the mineralogical changes due to long-term CO 2 -brine-rock reaction result in a roughly similar increase in frictional strength and a decrease in frictional stability with respect to the original OPA material (cf. Table 2, and Compared to brine-rock interactions, it can be inferred that, in the presence of CO 2 -rich brine, the extent of reaction-producing secondary minerals, with frictional behavior significantly different from the primary E Bakker et al. Original Research Article: Chemo-mechanical behavior of clay-rich fault gouges minerals, was not significant enough to impact frictional behavior drastically. Similar chemo-mechanical behavior has been seen in gouges derived from reservoir rock taken from natural CO 2 fields (unexposed and CO 2 -exposed Entrada Sandstone, Green River, UT, USA 35 and Röt Fringe sandstone, Werkendam, Netherlands.) 36 These observations suggest that the initial mineralogy of the fault gouge (and local variations therein) imposes a stronger control on fault behavior than the extent of CO 2 -brine-rock interactions, even after multiple pore volume flushes.
Furthermore, our geochemical simulations have shown that mass loss in an open fault system leads to significant porosity changes within the gouge (see Appendix D of the supplementary material). Such high gouge porosity will probably not be maintained over time. Compaction of the gouge will result in relaxation of the stress state in the surrounding host rock. Assuming a fully elastic response of the host rock and taking realistic values for the Young's modulus (in the range 2.7-8.2 GPa), 64 this stress relaxation will be concentrated in the first few centimeters surrounding the fault plane for depths relevant to CO 2 storage (>1.5 km depth).
What is evident is that ductile minerals, such as carbonates and sulfates, can potentially lead to velocity-weakening behavior (i.e., unstable sliding), particularly at elevated temperatures. 23,26,34 Carbonates exhibit velocity weakening under both dry and wet conditions (T > 75-80°C 23,26,35 ). Furthermore, studies have shown that swelling clay minerals, like smectite, are known for their very low frictional strength (μ = 0.1-0.3, see montmorillonite data in Fig. 6B-C 18 ). The presence of smectites, or precipitation thereof due to CO 2 exposure, can therefore lead to a reduction in the frictional resistance of the fault, making fault reactivation easier.
Although the impact of carbonates and/or smectites on friction and fault stability cannot be assessed accurately, it has been shown that the presence or formation of these minerals needs to be taken into account during site selection. However, such minerals would have to be present in quantities in excess of 20-40% to significantly impact fault behavior. 19,35 Such quantities would be difficult to achieve by CO 2 -brine-rock interactions alone given the low reactive capacity of most typical rock types considered for CO 2 storage sites. It should be noted that, with displacement, an internal fabric with throughgoing foliation might develop, potentially controlling the frictional behavior. Such a foliation may consist of a frictionally weak mineral and it could require as little as 4 wt% of material to create this weak-mineral foliation. 40 By contrast, in the event that carbonate-rich fluids, such as those produced by (local) carbonate dissolution, migrate up-fault and depressurize, CO 2 outgassing may take place, leading to large-scale carbonate precipitation impacting fault behavior (see also Bakker et al.). 35 Successful and safe CO 2 storage also requires maintenance of the sealing integrity of faults in the caprocks overlying storage reservoirs. Understanding of permeability change of the gouges with progressive CO 2 -induced alterations is needed to evaluate the sealing potential of faults, which was not investigated in the current experiments.

Conclusions
We assessed how the chemical-mechanical effects of long-term CO 2 -brine-rock interactions affect fault frictional behavior and stability in a clay-rich caprock. We employed Opalinus claystone as an analogue for clay-rich caprocks overlying typical reservoirs considered for CO 2 storage in Western Europe. Using geochemical modeling, we predicted mineralogical changes within the fault rock at timescales not achievable in the laboratory (10 000 years to 1 million years; 10-10 6 pore volume flushes). Subsequently, the frictional behavior of selected, evolved fault gouges was r The frictional behavior of simulated fault gouge, prepared from natural Opalinus claystone, shows friction coefficients of ß0.35-0.42 for temperatures ranging from room temperature to 150°C. Under all conditions, the gouges show velocity-strengthening behavior, meaning stable sliding upon slip acceleration. Modeled reaction with CO 2 -free reservoir water shows little variation in mineralogy but measured friction coefficients of the accordingly prepared gouge mixtures are ß10-30% higher than for natural OPA while the sliding behavior continues to be stable.
r Simulations of long-term (10 000 year and 1 million year) interaction between OPA fault gouge and CO 2 -saturated brine (10-10 6 pore volumes) predict dissolution of micas (up to 5 wt%), a reduction of carbonate content (up to 5 wt%), and an increase of quartz (up to 20 wt%) and clay content (up to 20 wt%). These mineralogical changes are significant compared to brine-rock reactions, such as the conversion of illite to kaolinite, but they still do not significantly change fault frictional behavior. Our findings are consistent with experiments performed on unreacted and CO 2 -exposed simulated fault gouges derived from material obtained from natural CO 2 reservoirs, such as the Entrada Sandstone (UT, USA) and the Röt Fringe Sandstone (Netherlands).
r Initial mineralogy of the fault gouge, and local variations thereof, imposes a stronger control on clay-rich fault behavior than the extent of CO 2 -brine-rock interactions. For fluid-carrying components that may promote precipitation of substantial amounts of secondary minerals with drastically different frictional behavior, such as (swelling) clay minerals leading to an evolution from a clast-supported framework to a phyllosilicate-dominated matrix, or frictionally unstable minerals, like carbonates, the impact on frictional behavior is obviously more significant. As such, the ratio between quartz and carbonate plays a role in controlling frictional behavior, due to the promotion of thermally activated processes, leading to ductile behavior in the more intensely crushed brittle carbonate grains. However, in most typical CO 2 storage sites, a source of chemically different fluids leading to substantial mineralogical changes may be limited.
r Overall, mineralogical changes due to CO 2 -brine-rock interactions can impact the frictional behavior and seismogenic potential of faults, and this can be assessed by the approach tested in this paper. However, it should be kept in mind that shear velocity, effective normal stress, and fault maturity, and internal fabric also remain important parameters, which cannot be neglected when assessing fault strength and stability.

Elisenda Bakker
Dr Elisenda Bakker is an energy and environmental consultant at Royal Haskoning DHV, where she gives advice on the physical and chemical processes that affect the mechanical and transport properties of rocks in the subsurface. She obtained her PhD from Utrecht University in 2016.

John Kaszuba
Dr John Kaszuba is an SER associate professor of geology and geophysics at the University of Wyoming. He is a recognized expert in high pressure/high temperature hydrothermal and multiphase fluid-rock reactions, integrating laboratory and computational approaches.

Sabine den Hartog
Dr Sabine den Hartog is a Lyell Research Fellow at Heriot-Watt University, Edinburgh. Her research focuses on the microscale mechanisms controlling fault frictional behavior, relevant to enhanced geothermal systems, carbon capture and storage, energy storage, and hydrofracturing operations.

Suzanne Hangx
Dr Suzanne Hangx is an assistant professor at the High Pressure and Temperature Laboratory at Utrecht University. She studies the physical and chemical processes controlling rock mechanical behavior in the subsurface in the context of CO 2 /energy storage and geoenergy production.