Microstructure and texture evolution in polycrystalline ice during hot torsion. Impact of intragranular strain and recrystallization processes

hot torsion. Impact of intragranular strain and recrystallization processes. Baptiste Journaux1,2, Thomas Chauve2,3, Maurine Montagnat2, Andrea Tommasi4, Fabrice Barou4, David Mainprice4, and Léa Gest2 1NASA Astrobiology Institute, Department of Earth and Space Sciences, University of Washington, Seattle, USA 2Université Grenoble Alpes, CNRS, IRD, G-INP, IGE, F-38000 Grenoble, France 3PGP, Department of Geoscience, University of Oslo, Norway 4Géosciences Montpellier, Université de Montpellier/CNRS, F-34095 Montpellier, France Correspondence: Baptiste Journaux (baptiste.journaux@gmail.com), Maurine Montagnat (maurine.montagnat@univ-grenoble-alpes.fr)


Introduction
Ice deforms by shear in many natural conditions such as glaciers and ice sheets, and in particular along ice streams. In most of the deep ice cores studied, compression and extension are the dominant deformation mechanisms owing to the cores location close to geographical domes or along ridges. Simple shear occurs in the deeper portions of these cores, but it is stronger along fast flowing glaciers and ice streams (see review by Hudleston (2015)). During large scale ice flow, deformation induces 5 development of strong crystallographic preferred orientations (CPO) or texture. Since the late 70's a growing number of studies have provided measurements of increasing accuracy of the texture evolution along ice cores (refer to Gow and Williamson (1976); Alley (1988);Lipenkov et al. (1989) for pioneer work, and Faria et al. (2014b) for a review). Pioneer work of Hudleston (1977) analyzed the texture evolution in a narrow shear zone of the Barnes Ice Cap (Canada), providing a well-documented observation of texture development owing to shear in natural conditions. Mantle rocks also develop strong olivine textures, 10 due to deformation and dynamic recrystallization (cf. reviews by Ismaïl and Mainprice (1998); Tommasi and Vauchez (2015)).
These textures give rise to anisotropy of both elastic and viscoplastic properties. The former is sampled in the mantle by seismic waves (seismic anisotropy) and the latter plays an important role in the development of plate boundaries (Tommasi et al., 2009) and in basal ice folding, for instance (see e.g. Dahl-Jensen et al. (1997); Thorsteinsson and Waddington (2002); Bons et al. (2016)). 15 Texture development in ice is related to the strong viscoplastic anisotropy of this material (Alley, 1992). Ice Ih is hexagonal with Space group P 6 3 /mmc, No.194 and Laue class 6/mmm. Plastic strain in ice Ih is mainly accommodated at high temperature by dislocation glide on the basal plane through three 1/3 <1120>(0001) equivalent slip systems (Duval et al., 1983;Hondoh, 2000). The viscoplastic anisotropy of the ice crystal results in strong strain heterogeneities at the inter and intra-granular scale when polycrystalline ice is submitted to a macroscopic stress (Duval et al., 1983;Grennerat et al., 2012) 20 (and reference therein) and hence in activation of dynamic recrystallization processes, even at very low strains (Duval, 1979;Duval et al., 1983;De La Chapelle et al., 1998;Chauve et al., 2017a).
In return, the texture-induced anisotropy at the macro-scale leads to a bulk anisotropic viscosity of ice that impacts ice rheology. This macro-scale mechanical anisotropy is responsible for modifications in ice stratigraphy and age-depth relationship (see e.g. (Martín et al., 2009;Buiron et al., 2011)). It may also produce large scale flow heterogeneities, such as basal folding 25 (see e.g. Dahl-Jensen et al. (1997); Gillet-Chaulet et al. (2006); Ma et al. (2010)). In the case of simple shear the evolution of texture can increase the shear strain by a factor of ten at constant stress (Cuffey and Paterson, 2010;Hudleston, 2015).
In addition to the obvious implications of ice rheology studies to ice caps and glaciers dynamics, ice is also often used as a reference material to study anisotropic mechanical behaviors in rocks and alloys. Since ductile deformation experiments in rocks are challenging, ice is a good analogue to polycrystalline geo-materials for which "high temperature" deformation can 30 be achieved in a cold room and ambient pressure, but at T /T m close to one. For instance, mantle rocks deformed close to the melting point in the asthenosphere beneath mid-ocean ridges, sampled in ophiolites, display strong olivine texture due to deformation and dynamic recrystallization with fast grain boundary migration (e.g. Cassard et al. (1981); Boudier and strain rate, or stress. This hypothesis has been confirmed by the analysis of natural samples deformed at low strain rates and high shear strains in a glacial shear zone (Hudleston, 1977). Surprisingly, the present state of knowledge on the microscopic processes occurring in polycrystalline ice under simple shear deformation is still limited to these data published over 30 years ago (Kamb, 1959(Kamb, , 1972Duval, 1981;Bouchez and Duval, 1982;Burg et al., 1986).
The tools and methods used then to analyze the texture were often manual and highly dependent on the operator experience. 15 Modern analytical tools are automated and more precise. Electron Back-Scattered Diffraction (EBSD) and Automatic Ice Texture Analyzer (AITA) can now provide high spatial and angular resolution quantitative data, enabling a global and statistical study of the processes accommodating strain at the micro-scale. A modern and comprehensive update of the observations of ice deformation in simple shear seems also more and more necessary as the increase of calculation power allows to run meanfield or full-field modeling routinely (see e.g. Montagnat et al. (2014) for a review). These new numerical approaches, aiming 20 at taking into account all strain accommodation mechanisms, require accurate experimental constraints on the microstructure evolution. It is worth noting that these models are still unable to accurately reproduce texture and evolution of polycrystalline ice and minerals under simple shear. For instance, the recent work by Llorens et al. (2017), which includes some recrystallization processes, does not succeed to reproduce the double sub-maximum of the c-axis preferred orientation or the interlocking grain shapes observed in (Bouchez and Duval, 1982). 25 The complex interplay of dislocation slip system activities, their interactions, and the role of the various mechanisms accommodating the local strain and stress heterogeneities, such as cross-slip, kink-banding, and dynamic recrystallization, is still not or poorly represented in most of the existing modeling approaches, very likely because of the complexity of the interactions between these mechanisms . Concerning the mantle, recently Signorelli and Tommasi (2015) succeeded to reproduce the evolution of olivine CPO in simple shear with a mean-field modeling approach, by making use of a two-level mechanical 30 interaction scheme that integrates the formation of low-angle grain boundaries as expected during dynamic recrystallization.
The present work aims at giving a modern update on our understanding of simple shear in polycrystalline ice by providing a comprehensive description of the emergence and evolution of texture and grain shape fabric in polycrystalline ice during hot torsion experiments. We use state of the art analytical techniques in material analyses: EBSD equipped with a cryostage and an Automatic Ice Texture Analyzer (AITA), to retrieve quantitative data of unprecedented spatial and angular resolution 35 3 The Cryosphere Discuss., https://doi.org/10.5194/tc-2018-213 Manuscript under review for journal The Cryosphere Discussion started: 7 November 2018 c Author(s) 2018. CC BY 4.0 License. on crystallographic orientations, grain size, and grain shape fabrics as a function of shear strain. These data then are used to estimate the active dislocation systems and to investigate the role of sub-grain scale processes on the recrystallization regime that allows the formation of a new texture and microstructure more favorable to shear strain.

Experimental methods
Polycrystalline ice preparation 5 Unstrained equiaxial polycrystalline ice was prepared by packing evenly 200 µm sieved ice particles in a mould. The container was sealed and the air contained in the porosity was removed using a primary vacuum rotary vane oil pump. Outgassed water at 0 • C was quickly introduced afterwards to fill the porosity. The resulting mush was frozen bottom up at -5 • C for 48 hours.
To enable a more homogenous grain size and recovery of any small strain accumulated during grain growth, the granular ice cylinders were annealed for at least 120 h at -7 • C. This method produces untextured homogenous samples with negligible 10 porosity and ∼1.5 mm mean grain size. Ultra pure Milli-Q™ water (18 MΩ·cm) was used for all sample preparations. Cylinders around 35 mm in diameter and 60 mm in height were carved from the granular ice using a lathe located in a -10 • C cold room.
Specific sample sizes are indicated in Table 1.

Torsion experiments
Deformation experiments were performed using a torsion apparatus placed inside a cold room at a temperature of -7±0.5 • C 15 (0.97·T m , with T m the melting temperature at 0 • C ). Samples were carefully fixed in the apparatus ensuring no axial load.
Straight lines parallel to the rotation axis were traced on the surface to control visually any strain heterogeneity along the cylinder (Fig. 1). The surface of the sample was then coated with silicon grease and wrapped gently with a cellophane foil to ensure negligible loss by sublimation.
A constant maximum tangential stress was applied on one side of the sample. We imposed a constant torque M corre-20 sponding to a maximum stress τ max between 0.4 and 0.6 MPa ( Table 1). The τ max was calculated from the torque M using a stress-strain power law as described in Paterson and Olgaard (2000), with a stress exponent of n = 3 (Duval et al., 1983). This value of maximum tangential stress has been chosen because it allows high enough strain rates, while preventing the formation of cracks. The actual torque applied value corresponding to each sample is indicated in Table 1.

Analytical methods
Texture, microstructure, and sub-grain strain heterogeneity analyses were performed using both AITA and EBSD, as the combination of the two techniques allows the acquisition of a comprehensive dataset. AITA enables the analyses of large samples (up to ∼10x10 cm 2 ), but only recovers the c-axis (optical axis) crystallographic orientations. On the other hand, the EBSD setup used is only capable of mapping smaller ice samples (∼1x2 cm 2 ), but is able to recover the full orientation of the crystal 5 (all crystallographic axes) with an expected angular resolution better than 0.7 • (Randle, 1992) and with a practical spatial resolution for ice down to 5 µm . Sections tangential to the experimental cylinders were cut out for both AITA and EBSD analyses. The general geometry of these sections is illustrated in Fig. 2. To provide flat surfaces for AITA thin sections and EBSD samples, the sample sections were microtomed by less than 1 mm, which gives an error on γ max of less than 5 %.
AITA has been performed on thin sections typically 300 µm thick, as it is routinely done on ice (Russell-Head and Wilson, 10 1999; Peternell et al., 2011;Chauve et al., 2015;Wilson et al., 2015). We worked at a spatial resolution of 25 µm and were able to obtain an angular resolution of about ∼ 3 • . All optical texture measurements were operated at -7 • C.
Grains boundaries (GB) were extracted using the segmentation technique described in Montagnat et al. (2015) from automated grain boundaries using color change detection in AITA maps, and manual corrections based on coherent microstructure shapes and discernible misorientations. All manual corrections are operator dependent, so all analyses have been performed by the first 15 author for consistency. This allowed us to differentiate, with a good level of confidence, sub-grain boundaries (SGB) from grain boundaries, even if optical measurements do not give access to the full crystallographic orientation, as EBSD measurements does. From the extracted skeleton images we derived quantified grain size and shape statistics using the PolyLX Matlab ® toolbox developed by Lexa (2003). From the grain shapes, we calculated the PARIS grain shape factor to quantify interlocking (angularity). It is defined by PARIS = 2·(P-PE)/PE*100%, with P the perimeter and PE the convex hull perimeter (Heilbronner 20 and Barrett, 2014). This parameter increases with increasing grain boundary sinuosity.
Electron backscattered diffraction (EBSD) patterns were obtained with a CamScan X500FE CrystalProbe equipped with a liquid Nitrogen cryogenic stage (GATAN ® ) at Geosciences Montpellier (France). Samples surfaces were prepared by carefully flattening the surface with a microtome blade in a -60 • C freezer just before loading it on the cold stage. This method provides a very good surface finish and limits to a minimum the heat transfer to the sample, ensuring limited microstructure evolution 25 due to annealing. The CrystalProbe-EBSD working conditions were 15 kV, 3.5 nA, and low vacuum with 1 Pa of gaseous Nitrogen and 25 mm working distance. More details on the EBSD geometry can be found in Demouchy et al. (2011). To avoid ice samples sublimation in the vacuum chamber, which happens at -60.6 • C at 1 Pa for ice Ih (Fernicola et al., 2011), the temperature of the cold stage was kept at −100 ± 10 • C during the analyses. EBSD Kikuchi patterns were automatically indexed using the HKL Channel5 ™ software suite (Oxford Instruments ® ). Maps were acquired with a spatial resolution of 30 25 µm as it was identify as a good balance between resolution and time constrains, with an indexation rate higher than 85%.
Analyses of the EBSD maps were made using the MTEX Matlab ® toolbox (Hielscher and Schaeben, 2008;Bachmann et al., 2010Bachmann et al., , 2011Mainprice et al., 2014). To observe the geometry of sub-grain boundaries and extract statistical quantities, local misorientation analyses, such as Kernel average misorientation (KAM) or the local misorientation relative to the mean orientation (Mis2Mean), are usually performed on EBSD data (see review by Wright et al., 2011). Both analyses enable a visualization of the spatial distribution of intragranular stored strain structures, generally in the form of sub-grain boundaries (Wright et al., 2011), but do not provide any information about the type of dislocations accommodating the observed misorientation. On the other hand, the Weighted Burgers Vector (WBV) analysis allows to extract a minimum value for the dislocation density in sub-grain boundaries and provides an estimate of the slip systems necessary to explain the observed misorientations (Wheeler et al., 2009). As conventional EBSD maps are 2D, only 5 components of the Nye tensor α can be extracted (α 12 , α 21 , α 13 , α 23 , α 33 ). This allows to obtain a projection of the Nye tensor on the EBSD map plane which constrain partially the Burgers vector. Hence it provides a lower bound for the dislocation density and the type of dislocations involved in the geometrically necessary dislocations of the WBV. A high-pass value of 0.4 • (||WBV||> 2.8 · 10 −4 µm −1 ) was imposed for the misorientation analysis to filter the noise resulting from the angular resolution of the EBSD data. We suggest to the interested readers to refer the to Appendix A of Chauve et al. (2017b) for more details about this analysis. Nevertheless, it should be reminded that the WBV analysis 20 does not directly provide information about mobile dislocations, responsible for the majority of the plastic deformation. It only measures the GND, as it is based upon the Nye tensor.

Results
Five torsion experiments were operated up to different finite shear strains γ max , from 0.012 to 1.94. Details for all experiments are indicated in Table 1. One sample (TGI0.71), with a γ max of 0.71, was annealed for 72 hours at -7 • to study the effect 25 of annealing on both texture and microstructure. The microstructure of one of these samples has been partially described in previous work of our team to compare the statistical representation of WBV (or Burger vector) with [c]-component in GNDs in samples submitted to uniaxial unconfined compression and to torsion (Chauve et al., 2017b).

Macroscopic strain
Retrieved samples presented homogeneous strain with no apparent localization or cracks, as illustrated by the photographs in  Fig. 1 until approximately 20,000 s (≈ 6 h).The noise in TGI0.012 and TGI0.2 data was too strong to distinguish any primary creep hardening. After this phase of hardening, a minimum in strain rate is reached (secondary creep regime).

TGI1.96 in
This minimum is clearly visible in experiments TGI0.2, TGI0.42, TGI0.71 and TGI1.96 at 200,000 s (≈ 56 h). For the longer TGI1.96 experiment, after the secondary creep phase, strain rate continuously increases until achieving an almost steady state strain rate of 8.5·10 −7 s −1 , suggesting that tertiary creep regime was reached for this run (Fig. 1). TGI0.012 stayed mostly in 5 the primary creep regime with a γ max = 0.012. Samples TGI0.2 and TGI0.42 experienced a maximum shear strain of 0.2 and 0.42 respectively. The final strain rate for these two experiments is illustrated by the tangent dashed line in Fig. 1 which shows that softening was still occurring when the samples were unloaded (only the last data points lies on the tangent). We therefore consider that those experiments where still in a transient regime when stopped. A strain step is visible for TGI0.42 between 550,000 s (≈ 153 h) and 650,000 s (≈ 180 h) ; it corresponds to a period where the weight used to apply constant torque was 10 in contact with the metallic frame of the torsion apparatus, consequently decreasing the applied stress. After 650,000 s (≈ 180 h), the weight was put back in a good configuration and the strain curve recovered its pre-step slope. The general variation in strain rate evolution with time between the different runs is due to the variability in grain size and textures in the starting material, slightly different sample sizes, and variations in the applied torque (Table 1).

15
AITA maps of tangential sections for each sample and associated analyses are presented in Fig. 2 and Table 2. The evolution of texture, microstructure (grain size and shape) as a function of finite shear strain is clearly visible. The color wheel gives the orientation of the <0001> c-axis. The annealed sample TGI0.71 is discussed separately at the end of this section. As the shear strain increases, the distribution of colors changes from random color toward a predominance of red and blue grains, until the map becomes dominated by red color grains for TGI1.96 (γ max =1.96). 20 This evolution records the progressive change in orientation of the c-axis with increasing shear strain, displayed as pole figures in Fig. 2b. For the undeformed ice sample (γ max =0) we observe random initial orientations of c-axes with a texture index J i = 1.07. The texture J-index is derived by calculating the second moment of the Euler angles orientation distribution function (ODF) of discrete crystal orientation data (Bunge, 1982), and is commonly use to quantify texture strength in geology and in material science. The strain reached by TGI0.012 sample is not high enough to produce any significant change of its 25 texture. As the strain increases we observe the formation of two sub-maxima, as reported in previous studies (Kamb, 1959(Kamb, , 1972Duval, 1981;Bouchez and Duval, 1982;Burg et al., 1986), one almost perpendicular to the shear plane (M1) and the other one at high angle (45-90 • ) from the shear plane (M2) as indicated in Fig. 2. With increasing shear, the maxima M1 and M2 increase and decrease in intensity, respectively. This evolution towards a single maximum CPO results in an increase of the J-index (Fig.3). The plane containing the finite extension direction (ED) and normal to the compression direction (CD) has 30 a very low [c]-axis CPO density for γ ≥ 0.2. One should note that for 0.2 ≤ γ ≤ 0.42, the bimodal pattern formed by the two M1 and M2 submaxima seems to be symmetric relatively to the maximum finite stretching axis. With increasing shear strain the angle between M1 and M2 decreases, but the symmetry is maintained. AITA extracted grain statistics are presented in Table 2. With increasing γ the median grain size diminishes from 0.7(4) mm to 0.12(16)mm. The median is preferred relative to the mean because of the non-normal distribution of grain sizes. The coarser grain size in TGI0.42 sample compared to TGI0.2 and TGI0.012 is probably due to a difference in the initial unstrained grain size, but the grain sizes remain comparable if interquartile ranges are considered. Grain shape fabrics were calculated using moment-based ellipse fitting technics (Lexa, 2003), results are illustrated as rose diagrams in Fig. 2c. As the shear strain Grain boundary interlocking estimated through the PARIS factor (reported in Table 2), seems to increase during the transient strain regime (from 2.2 % at γ = 0 to 11.5 % at γ = 0.42). It seems to decrease at a high shear strains of γ = 1.96 (PARIS = 5.9 % ).
The annealed sample TGI0.71 shows a median grain size of 1.7 mm, above the initial unstrained value of 0.7 (Table 2). The CPO seems not modified by the annealing with the M1 and M2 maxima maintained and a J-index = 2.96, which lies in between 5 the values for γ max 0.42 and 1.96, as would be expected from a non-annealed sample with the same finite shear strain (Fig.   2b). The grain shape fabric in the annealed sample is nevertheless more marked than in the strained, but non-annealed samples.
The PARIS factor decreases significantly, as much of the interlocking of grains is smoothed by annealing.
As a summary, the crystallographic orientation evolution during simple shear strain of polycrystalline ice can be described from our data as follows. M1 orientations rapidly develop (γ = 0.2) and increase in intensity until they become predominant at high strains (γ > 1.96). The M2 maximum also rapidly develops symmetrically to M1 respectively to the maximum finite strain direction. However the M2 maximum continuously decreases in intensity during the transient regime. The other crystallographic orientations start to disappear as early as γ = 0.2 and this tendency is accentuated with increasing shear strain.
Finally, if annealing processes occurs after the strain stops, the CPO is maintained even if the grain shape fabric and grain size are modified.

Maps & textures
Multiple EBSD maps were acquired for every final shear strain. Representatives examples of EBSD maps for each sample are shown in Fig. 4.a, with the number of maps n m acquired for each γ max indicated at the bottom. Spatial resolution is of 25µm.
The access to the full crystallographic orientations allows to perform accurate grain segmentation based on misorientation data. The orientations of <0001> <1010> and <1120> axes relatively to the shear plane and direction are shown in Fig. 4.b.
All maps acquired for each γ max were combined for better statistical analysis. An excellent level of agreement with CPOs from AITA is observed for <0001> textures, which develop rapidly M1 and M2 sub-maxima, with disappearance of the M2 maximum and reinforcement of the M1 one at high shear strain. The increase in texture J-index confirms this evolution, with 15 very consistent values with those based on the AITA measurements for a same γ max (Fig. 3). Crystallographic orientations of <1010> and <1120> both evolve during strain, until they form a girdle along the macroscopic shear plane at γ max =1.96. We do not observe a significant difference in orientation between <1010> and <1120> at any shear strain.  ice and in natural environments, Mansuy et al., 2000;Piazolo et al., 2008;Weikusat 5 et al., 2011;Faria et al., 2014a;Chauve et al., 2017b). As described in the methods section, WBV analysis allows to derive from intragranular misorientation maps the accumulation of GNDs and to define the relative proportion of dislocations with < a > and [c]-component Burgers vector within GNDs. Figure 5 provides representative ||WBV|| maps at every finite shear strain.
In these WBV maps (Fig. 5) we observe a large number of linear sub-grain boundaries already developed at γ max = 0.2.

10
Evolution of the GNDs density with increasing strain is hard to assess by comparing visually the maps for TGI0.42 and TGI1.96, but it can be assessed by the analysis of the quantitative statistical distribution of < a > and [c] GNDs (Fig. 6).  (Piazolo et al., 2008;Weikusat et al., 2011;Piazolo et al., 2015;Chauve et al., 2017b). Such subboundaries are also identified in the present study (green to yellow lines in Fig. 5b) It is interesting to note that even the most deformed sample, which has a strong CPO with one single sub-maxima at γ max = 1.96 (at steady state tertiary creep), GNDs 10 are very common. In the almost undeformed (TGI0.012) and annealed cases (TGI0.71), almost no sub-grain boundaries are identifiable.
As demonstrated by Chauve et al. (2017b), the total density of GNDs with a significant ||WBV|| clearly increases with shear strain (Fig. 6f). However, the ratio of We computed the probability density function of rW BV c as function of the distance to the grain boundary, and represented it for rW BV c < 1/3, 1/3 < rW BV c < 2/3 and rW BV c > 2/3 in Fig. 7. We are able to observe that the rW BV c < 1/3 density 25 clearly decreases around 100 µm close to the grain boundary, when both 2/3 < rW BV c < 2/3 and rW BV c > 2/3 densities increases (rW BV c > 2/3 almost doubles). This statistical analysis clearly shows the increase of [c]-component dislocations proportion in the substructures at a distance lower than 100 µm from the grain boundary. It is also important to note that we do not observe any significant differences between different finite shear strains.

30
The present data corroborates the previous observations on the texture evolution in ice deformed by simple shear by Hudleston (1977); Bouchez and Duval (1982), with an important increase in quantitative analysis and pushing the statistical analysis down to the intragranular level. As indicated by the evolution of the macroscopic strain rates on Fig. 1, we were able to analyze the texture and microstructure at different creep regimes. The near constant strain rate (ε f = 8.5 · 10 −7 s −1 ) achieved for the TGI1.96 run can be interpreted as the evidence that this sample was close to steady state in terms of mechanical behavior, typical of tertiary creep regime. The apparition and evolution of c-axis orientation sub-maxima M1 and M2 in the texture are consistent with previous data from Hudleston (1977) and Bouchez and Duval (1982). The M1 maximum remains always normal to the shear plane and the M2 5 sub-maximum, initially as strong as M1, progressively rotates with shear during transient creep, and decreases in intensity until it disappears, possibly as tertiary creep regime is reached.
The slow evolution of the grain shape preferred orientations and their saturation at shear strains higher than 0.2, visible from the grain shape orientation rose diagram (Fig. 2.c), could be the effect of recrystallization processes and fast grain boundary  migration that counters the formation of a strong grain shape fabric. These processes may also explain why we observe a decrease in the grain boundary interlocking at high strains and even in the annealed sample through the PARIS factor (Table   2). Indeed nucleation should enable the apparition and growth of new grains with small PARIS factors at the expenses of older grains with serrated grain boundaries (i.e. higher PARIS). This observation underlines that the estimation of strain magnitude from grain elongation or shape is not reliable at high temperature, as noted by Burg et al. (1986). The efficiency of grain 5 boundary migration and recovery of shape fabric is expected to significantly decrease with decreasing temperatures. The case of the annealed sample TGI0.71 is interesting to understand how metadynamic annealing processes like grain boundary migration driven by dislocation density variation may change the microstructure and texture (CPO). The texture is maintained, the grain shape fabric is reinforced, and the final grain size is larger than the initial one. The preservation of the texture suggests that most new "dislocation free" nucleii were already close to M1 or M2 orientations. This would favor predominance of nucleation 10 by bulging over spontaneous nucleation, as the later tends to produce random orientations (Chauve et al., 2017a). This also means that if stress is somehow lowered in natural context, and annealing takes place, the texture could still be interpreted, even if information about grain shape fabric is lost. The reinforcement of the grain shape fabric visible on the rose diagram in Fig. 2.c also suggests some oriented grain growth. It is possible that during strain, the formation of a large number of isometric nucleii weakens the shape fabric, but that this signal is erased during static annealing. 15 The full lattice preferred orientations derived from EBSD data reveal that neither <1010> nor <1120> axes display any preferential orientation within the shear plane, even at high shear strain (γ = 1.96, Fig. 4). This corresponds to an isotropic plastic response on the basal plane during simple shear strain in polycrystalline ice. 1/3 <1120>(0001) slip systems are supposed to be dominant in the basal plane (Hondoh, 2000); this assumption is consistent with the present data, but glide in the <1010> direction is also possible. There have been also direct observations of GND loops in ice that require [c]-component dislocation walls to be closed (Chauve et al., 2017b). This would suggest that the increase of [c]-component GNDs near grain boundary helps to form GND loops that give birth to new nucleii. These observations, combined with our present WBV statistical analysis, point towards a possible way 30 to estimate the maximum size of bulging nucleus, which would be lower than 100 µm in the present case. It would be important in future studies to test if this critical dimension would vary with temperature or stress, as well as to obtain misorientation data at higher spatial resolution to better constrain the evolution of this higher [c]-component dislocation zone. The fact that the possible maximum size of bulging nucleus is constant at all finite shear strains suggests that bulging is very active at all creep regime, further stressing its importance for understanding the rheology of ice and other materials at high temperatures. Our experimental results and the ones of Bouchez and Duval (1982) are in general disagreement with modeling of shear in ice either by mean-field approach as in (Wenk et al., 1997) or more recently by full-field modeling as in (Llorens et al., 2017).
This later work predicts only one single submaxima inclined from the tangent to the shear plane, and a preferred orientation of <1120> within the shear plane. These models consider that plastic strain from dislocation slip results not only from slip on the basal plane, but also on prismatic and pyramidal planes. By activating these later slip systems, the lattice is geometrically forced to rotate away from the normal to the shear plane, creating unrealistic textures neither M1 nor M2 sub-maxima. On the contrary, pioneering work on 2D modeling of polycrystalline aggregates under simple shear by Etchecopar (1977) was able to reproduce the 2 sub-maxima M1 and M2. This was simply done by considering a single slip system (basal slip system for ice) and allowing cells to subdivide (polygonization) and undergo rigid body rotation. The very good agreement of this simple model with evolution of textures observed experimentally for ice under shear was emphasized by Bouchez and Duval (1982), 10 who hypothesized that the polygonization processes in ice would be formation of GNDs and kink-bands. In our results few kink bands were observed, but the prevalence of GNDs at most finite shear strains suggests that Bouchez and Duval (1982) supposition is reasonable. Indeed, rotation of the dominant slip system into parallelism with the imposed macroscopic shear is predicted by the mean-field VPSC model modified to account for subgrain rotation recrystallization (Signorelli and Tommasi, 2015). Yet, in these study parallelism is only achieved at shear strains of 5, whereas in the present study and in Bouchez and 15 Duval (1982) a single maximum texture is observed for shear strains as low as 2. This points to an additional contribution to the texture reorientation, which at the high homologous temperatures of both experimental studies might be nucleation by bulging associated with sub-grain rotation as a bulge closure, and fast grain boundary migration.
The analysis of the texture evolution in the experiments suggests that plastic strain through dislocation creep is probably as commonly claimed, largely dominated by basal slip. Nevertheless, stored GNDs (including non-basal GNDs with 20 [c]-component Burgers vector) allow recrystallization accommodation mechanisms such as nucleation by bulging. The fact that the PARIS factor initially increases then decreases above γ max =0.42 enables to make further assumptions about active recrystallization mechanisms and their evolution (or sequence) with increasing strain. Until a strain close to γ max =0.42, heterogeneities in dislocation storage close to grain boundaries favor spatially irregular grain boundary migration to dominate, creating numerous serrated grain boundaries, as indicated by the increasing PARIS factor. With increasing finite strain, 25 the influence of nucleation gradually increases, forming more isodiametric grains, lowering the PARIS factor, and eventually erasing the M2 texture. Indeed, SGBs required to close bulges may necessitate strong local stress levels (in order to activate [c]-component dislocations, for instance) that are obtained after a given amount of accumulated strain, while grain boundary migration may occurs since the very beginning of the deformation.
Our results also underline the efforts remaining to match experimental observations and mean-field and full-field numerical 30 modeling. Comparing our data to Llorens et al. (2017) models, it seems clear that the numerical approach is not able to reproduce the initial 2 sub-maxima, and the later single maxima, normal to the macroscopic shear plane. This could be due to several processes that are not taken into account in the models and that, based on the present data, seem to play an important role, namely the bulging nucleation process and the possibility of non-basal dislocation slip to participate to SGB related polygonisation. 35 17 The Cryosphere Discuss., https://doi.org/10.5194/tc-2018-213 Manuscript under review for journal The Cryosphere Discussion started: 7 November 2018 c Author(s) 2018. CC BY 4.0 License. By using state of the art analytical techniques (i.e. AITA, EBSD and WBV analysis), we were able to characterize the texture, microstructure, and geometrically necessary dislocation structures in ice deformed under torsion in the laboratory at an unprecedented resolution. The experiments, performed at high temperature and up to shear strain of 2, favored dynamic recrystallization mechanisms, in conditions close to those encountered in cold glaciers, ice stream and even some deep ice core areas.

5
From these high-resolution observations, we were able to highlight the impact of grain boundary migration and nucleation by bulging, associated with sub-grain boundary formation, as the main mechanisms responsible for texture and shape fabric evolution with strain. Sub-grain boundaries density and dislocation components are characterized as abundant at all observed finite shear strain, even at this high temperature (0.97·T m ). Furthermore, around 30% of Geometrically Necessary Dislocations exhibit a significant [c] component, due to strong local stresses arising from the strong viscoplastic anisotropy of the ice 10 crystals. This [c] component increases close to grain boundary, suggesting that this process may efficiently participate to the closing of bulging grain boundary during bulging nucleation.
So far, full-field modeling approaches that simulate plastic deformation by activating non-basal slip systems in ice are unable to reproduce the strongly clustered simple shear texture observed experimentally and in the nature. As was already suggested by Etchecopar (1977), polygonization may play a crucial role to accommodate dislocation glide dominated by basal slip. 15 Our observations suggest that bulging, associated with sub-grain boundaries mechanisms, could be responsible for this and explain the observed strongly clustered texture at relatively high strain and high temperature. We conclude that that non-basal dislocation slip should not be involved in modeling to obtain realistic textures in polycrystalline ice and other anisotropic material, but that GNDs with a [c] components should play a major role in strain accommodation and nucleation processes. We hope that these new results will help the modeling community to provide realistic modeling texture evolution of simple shear 20 in ice, which are capital for a better predictive ability of ice landmasses evolution in the coming decades and centuries.
PolyLX is a free MATLAB ® toolbox for quantitative analysis of microstructures, developed by Lexa (2003), and available on https://petrol.