F'ractals, Vol. 2, No. 4 (1994) 567-581 @ World Scientific Publishing Company FRACTAL PLASTIC SHEAR BANDS ALEXEI N. B. POLIAKOV* and HANS J. HERRMANN, PMMH (U.R.A. 857), ESPCI, 10 rue Vauquelin, 75231 Paris, Cedex 05, France HLRZ, KFA Julich, Postfach 1913, 52425 Jiilzch, Germany YUFU YU. PODLADCHIKOV* Department of Sedimentary Geology, Vrije Universiteit, 1081 HV Amsterdam, The Netherlands STEPHANE ROUX PMMH (U.R.A. 857), ESPCI, 10 rue Vauquelin, 75231 Paris, Cedex 05, France Received September 12, 1994; Accepted September 16, 1994 Abstract We present a numerical study model of shear bands in rocks with a non-associated plastic flow rule. The system drives spontaneously into a state in which the length distribution of shear bands follows a power law and where the spatial organization of the shear bands appears to be fractal. The distribution of local gradients in deviatoric strain rate has different scaling exponents for each moment which we calculate and discuss. Samples of granodiorite from the Pyrenees sheared under high confining pressure are analyzed and their properties compared with the numerical results. 1. INTRODUCTION a network of thin shear bands crisscrossing each n i ~ l using Carrara Already in 1911 von ~ & ~ r showed marble t h a t when rocks are cornpressed under sufficiently high hydrostatic pressures and sheared, they undergo a transition from brittle fracture t o plastic deformation characterized by t h e appearance of other at roughly 45' with respect t o the ~ r i n c i ~ a l stress axis. IAter the same effect was shown for other types of rocks as sandstones, ~ i m e s t o n e as ~~?~ well as for soils and sand.4 There are also numerous observations of plastic shear zones in the field.5v6 "Also affiliated to the Institute of Experimental Mineralogy of RAN, Chernogolovka, Russia. 567 568 A . N . B. Poliakov et al. Fig. 1 Natural example of ductile shear bands in a rock (very homogeneous granodiorite) from the Central Pyreneas (region of Cap de Long). The 200 FF bill in the lower left corner serves as scale. The box-counting technique gives a fractal dimension of 1.58 f 0.08. Figure 1 shows dark conjugate shear zones that we obtained from a granitoid of the Neouvielle Massif in the Central Pyrenees consisting of a very homogeneous granodiorite with equiaxes minerals of an average size 1-2 mrn. This massif has been deformed during Alpine orogenesis, thus developing evenly distributed conjugate shear zones on different scales ranging from centimeters to kilometers. We present examples on the scale of several meters which we obtained in the central part of the massif along the Cap de Long. Recently a n analysis of granite using a petrographic microscope7 showed that a cut through a fault network gives fractal structures with fractal dimensions depending on pressures and the nature of the external load. We present in this paper a numerical simulation of the shear bands of elastoplastic materials on different scales and calculate their fractal properties. We also measure for the first time in the field the fractal dimension of shear bands in granodiorite formed under confining pressures of around 5 kbars. 2. CONSTITUTIVE RELATIONS It is well known from experiments that rocks increase in volume and dilate, when distorted by shear. Dilation occurs by tensile cracking and is necessary to lift sliding blocks over asperities. A parameter suitable for characterizing a dilatant material is the dilatancy angle $. In t he case of simple shear, tan($) is the ratio of plastic volumetric strain rate divided by plastic shear strain rate. Materials with $ = 0 are plastically incompressible and so conserve volume. Materials with higher $J dilate more with plastic strain. This gives the plastic flow rule which besides the yield criterion characterizes the type of plasticity. The plasticity of rocks is quite different from classical plasticity of metals,$ not only because of dilatancy but also because the yield linearly depends on pressure (Mohr-Coulomb criterion). The onset of plasticity is well described by a relationg: where T and cr, are the shear and the normal stresses resolved on any plane within the material and the material constants 4 and c are the "friction angle" and the "cohesion", respectively. In general, the flow rule of plasticity is non-associated, i.e., the angle of dilatancy $ under shear is different from $ - the plastic strain increments are not normal to the yield hyper-surface. The friction angle $ is typically 30-40' and the dilation angle is smaller than 10' for rocks. Fmctcsl Plastic Shear Bands 569 A stability analysis10 shows that in some cases lane strain) a n important consequence of nonassociativity is t h e localization of initially homogeneous deformations into shear bands even if the material strain hardens. Thus we obtain localization in t h e pressure-dependent plastic material due to a rheological instability without softening, heterogenities i n material properties or non-uniform boundary conditions. More details can be found elsewherea8>11 3. NUMERICAL MODEL We have used a n explicit Lagrangian technique similar t o FLAC as developed by Cundall12to simulate an elastoplastic two-dimensional medium under plane-strain condition. FLAC is a very powerful technique to simulate nonlinear rheological behaviors a t very high resolution due to an explicit time-marching scheme which does not require to store large matrices which are typical for implicit methods. We applied pure shear on a square sample in two different ways, either by prescribing fixed velocities at all boundaries (FBV), as follows13: or applying periodic boundary conditions (PB) keeping the spatial average of gradient of velocities t o a prescribed value: where a is a constant and i is a strain rate. Both of these boundary conditions give very similar and nearly uniform distributions of localization for systems having more than 100 x 100 elements. (Note t h a t t h e maximum velocity will be at the boundaries a n d will be denoted as Vb,in the following.) Besides the abovementioned material constants c, 4 a n d Q we can also change the elastic moduli or Lam4 constants X and p. All elements were initialized w i t h the same confining pressure p. 0 ur numerical technique has an intrinsic time step At chosen sufficiently small so that elastic waves propagate only a fraction 6 of the length of one element ensuring numerical stability of the explicit scheme: and found that the results do not depend o n this parameter. Since our initial setup was completely homogeneous the random nature of the shear bands is due to minute effects in the round-off of the floating point numbers in the computer. These round-off errors only trigger the initial position of the shear bands but have no influence on the further evolution of their pattern. We checked this by simulating a system with weak quenched disorder and found the same result as in the homogeneous case. The only difference between these two simulations w a s that localization in the homogeneous case started later than in the inhomogeneous case. This delay occurs because in the homogeneous case it takes some time for the round-off errors to accumulate sufficiently in order to trigger the instability. The results presented in this paper are shown after a transient time, sufficiently long to ensure that the average number of shear bands does not change anymore with time. The location of the shear bands and their strength is, however, not constant with time as we will discuss in the next section, An important point is worth being mentioned. In order to reach equilibrium, the algorithm includes a damping term. The latter rnimicks a viscous effect, but not with a physically fixed viscosity, but rather proportional to the out of balance force at each node, as in Cundall original code.12 Therefore, when the system is in equilibrium, the viscous term vanishes and thus one obtains the solution looked for. However, dynamic effects may be sensitive to the numerical damping. This point is to be considered as we shall see that the structure of t h e shear banding we will observe is essentially a dynamic effect dependent on the loading velocity. However, the algorithm used is so much more efficient for handling large size systems that a comparison with other methods over such length and time scales is merely impossible. Therefore, we report the numerical results obtained but we cannot reach a definite conclusion on the role played by the artificial damping. We however feel that the results we present below shows such a remarkable agreement w i t h the field measurements that the numerical results are worth being reported and analyzed in detail. 4. RESULTS where Vsound is the p-wave velocity and p is the density of t h e material. We performed several calculations w i t h different time steps (6 = 0.1, 0.25, 0.5) 4.1 Stress and Strain Rate Distributior Figure 2 shows a snapshot of the second invariants of (a) strain rate; (b) deviatoric stress; and ( c ) local 570 A. N. B. Poliakou e t al. rotations. We visualize in colors t h e local changes of the second invariant of the strain rate [Fig. 2(a)] such that in the yellow and red regions there are strong changes either in the direction or in the magnitude of the motion of the material. The blue regions are elastic. In Fig. 2(b) high deviatoric stresses are presented by red color and low ones by Fig. 2 Snapshots of the (a) strain rate; (b) deviatoric stresses; (c)local rotations for bi-periodic boundary conditions, using q5 = 40°, $J = OD, c = 0, X = p (Poisson ratio, u = 0.25), p / p = or B = p / p . Vsound/Vbc= 1. Fig. 3 Comparison of a numerical result (accumulated plastic strain) (a), with a natural rock from the Pyrenees (b). The scale of (b) is about 2 meters (A 5 FF coin can be seen slightly above the center of (a)). Fractal Plastic Shear Bands 571 blue color. In Fig. 2(c) local rotations are coded such that clockwise rotations are red, anticlockwise ones are blue, and green denotes essentially rotationless regions. In all cases the color scale has been chosen to match exactly the range of values obt ained in the picture, so that small inhomogeneities may be exaggerated. We see that spontaneously shear bands are formed in which the plastic deformation occurs. These bands form an angle 8 with 45' - $12 < B < (45O - q / 2 ) with respect to the horizontal which is consistent with bifurcation theory."$14 In Fig. 3 we compare the network of shear bands of natural granite with the results of a simulation Fig. 4 Strain rates maps obtained for different friction 4 and dilation .IC, angles. Note that localization is stronger with increase of non-associativity g5 .IC,. Boundary conditions are FBV, system size is 300 x 300, and the other parameters are the same as in Fig. 2. Color coding as in Fig. 2(a). - with the same parameters as in Fig. 2, except that the pressure p is increased by 100 (B = 100). If we identify one discrete element of our simulation with the size of a grain of the rock (R 5 mm), the samples of Figs. 3(a) and 3(b) have about the same size (the number of grains in the sample is of the same order as the number of elements in our numerical model). A difficulty in comparing the two pictures arises from the fact that we do not know the geological history of the rock so that on the one hand, we cannot compare the time scales, and that on the other, the natural rock is a cumulation of all shear bands that have occurred, and one cannot exclude that during the geological deformation, the external load changed direction or magnitude. Note that for comparison we do not show the snapshot of strain rate as in Fig. 2(a) but the accumulated plastic strain after l o 4 time steps. 4.2 Dependence on Parameters First let us discuss the influence of the dilation and friction angles on the localization. Figure 4 shows the strain rate for different 4 and $ and other identical parameters. Figures 4(a) and 4(d) demonstrate that a small difference of these angles (20") gives thick and regular shear bands. An increase of the difference between these angles enhances the localization and makes shear zones thinner and more irregular [see Fig. 4(b) with C$ -1C, = 40' and Fig. 4(c) with q5 - 3 = 60'1. It is interesting to note from comparing Figs. 4(a) and 4(d), that 4 - is not sufficient to determine the shape of the shear bands but that the absolute value of the angles is import ant. We performed many numerical experiments studying all possible parameters which can influence the evolution of the shear bands. The parameters of our model are: e * the elastic properties p and A (since we did not vary the Poisson ratio, one of these two parameters characterizes the elastic stiffness of the material X = ,u) the plastic properties which are as discussed above, gl, $ and c. In the rest of this study, c = 0, 4 = 40" and ll/ = 0'. the system size. Since there is no intrinsic length scale in the problem, only the ratio between the system size and the mesh size is relevant. We call this ratio L. the loading characteristics, which contains two elements. The initial hydrostatic pressure p and the imposed average shear rate. The latter is characterized by the velocity along the boundary Vbc. Written in dimensionless form, three parameters appears: the system size L, the ratio between the hydrostatic pressure and an elastic constant p / p which can be interpreted as proportional to the elastic deformation under the initial hydrostatic pressure, and finally the ratio of the sound velocity to the boundary velocity c/Vbc. All other parameters are kept constant in the rest of this work. In fact a useful combination of the latter pararneters can be introduced. The shear strain after a time t can be written as &,t/L. Thus the time at which the yield limit will be reached is r, = pL tan(+)/(p&,): The time for an elastic wave to travel through the system is T,, = Llc. The ratio of these two intrinsic time scales gives a non dixncnsional number which we define as B: where we have dropped the tan(&) termed which will be kept const ant. We found numerically that the shear band patterns only depends on B and L. Changing the pressure p, the shear rate hc, the sound velocity c or the elastic properties ,u, while keeping B unchanged does not affect the results. This is shown in Fig. 5. We see that the evolution of the system depends only on this non-dimensional parameter B by varying confining pressure, elastic modulus and velocity of loading. Figure 6 shows that the spacing of shear bands is controlled by this parameter. The decrease in shear band separation with increasing B can be qualitatively explained the following way: It is l ~ n o w n , that ~ ~ when ~ ' ~ a~band ~ ~ forms it causes a decrease of mean stress inside of the band due to elastic unloading. The pressure outside the band increases, inhibiting the formation of another parallel band nearby. This change in pressure propagates through the system at the sound velocity and thus at high B (large c), the band formation will affect a larger scale. On the contrary, at small B the sound velocity is so low compared with the loading rate that the shear bands have to be X X ~ O ~ I ? densely distributed to release the imposed shear. The dependence on the other properties p, p, . . . cam be analyzed in similar terms using the construction of B. In Fig. 7 we show how the average spacing between the shear bands as measured with a ruler Fractal Plastic Shear Bands 5 g. 5 Various parameters having the same value of B and nensionless control parameters. Color coding as in Ipig, 2(a from Fig. 6 depends on B. As we will see in the next section t h e spacing also depends critically on t h e resolution L (fractal character) so that it is important in Fig. 7 to work at constant system size. W e see t h a t t h e spacing increases - though slowly - with B , in agreement with the previous discussion. The quantitative analysis of this variation is .), showing that these two parameters are t h e only t~ to be taken with caution. Indeed, as can be seen from Fig. 6, when B is less than 1, the s h e a r bands form domains in which the bands are all parallel. There are almost no crossing of bands. For larger B, on the contrary, the two families of conjugated shear bands (with opposite slopes) are intermingled and do not form domains. This might explain t h e Fig;. 6 Dependence of the strain rate on the para]meter B = p / p . V lound/V~c , [or systems of size 300 x 300 and periocdic boulndary conditions. T h e other parameters are the slame as in Fig. 2 an d the color coding is like that in Fig. 2(a). Fractal Plastic Shear Bands 575 100 1 -4 10 1o - ~ 1 o" 1 o2 1 o4 1 o6 B Fig. 7 Log-log dependence of the average spacing A S between shear bands as a function of the parameter B, The result can be approximated by the power law as A S oc B'.~. apparent discontinuity of the band separation for small B in Fig. 7. For large B, one begins to be sensitive to the finite size of the system, and the effective separation might be significantly altered by the presence of periodic boundary conditions. Other boundary conditions are even more strongly biased because of the constraint of the imposed velocity close to the boundary. A denser distribution of bands arises in this case. However, for the practical purpose of using this separation, we can fit roughly a power-law of exponent 0.2 through the data. It is possible however that this apparent behavior is far from what it would be in an infinite system size. Another interesting question is to monitor the time evolution of shear bands. To study this problem, we make a vertical cut through a plot of strain rates as shown in Fig. 6 and record the evolution of this one-dimensional slice with time. Figure 8 shows a plot where the horizontal axis gives strain-rates along the cut and the vertical axis is time. Different time-dependent behaviors are observed depending on the parameter B. For low B there are regular and stable shear bands. For high B shear bands have a strong time-dependent behavior and a fluctuating intensity. The life time is also shorter for decreasing B. Again this emphasizes the fact that the structuration of the shear banding is essentially a dynamical effect of our model. Figure 9 shows the time evolution of the stress. Stress components (ox, and oxy) averaged over all elements. Both values can be calculated analytically for a single element (a,, = -p . (1 f sin($)) c cos($), os, = 0) and we see that in this sense the total system has the same global perfect-plastic behavior as the single element. On the contrary, the stress (pressure and maximum shear stress) in a single element is strongly fluctuating with time. a 576 A . N. B. PoliaLov et al. i r : i - {j.!/: Fig. 8 Time evolution of the shear bands for different B. T i ~ n eincreases vertically upwards. Color coding as in f p i g , l ( a ) . Because the material is closc/or a t the yiclcl, the pressure and the shear stress fluct.uation of a single element are autocorrelatecl with each other. Thus despite t h e overall rcsponse of the sample being sirnple and similar to the characlcrist.ic of a single elcment, there is a strong stress arld st,rain rate hcterogeneity in space and strong fiuct.nations with tilric for every single elcxncnt. 4.3 Fractal and Mult ifractal Analysis In Fig. 10 we see snapshots from the evolution of the system with different mesh sizes L but the same physical parameters. Evidently, 1)ccansc. of t.lio 1111dcrlying instability the shear bantls bocolllcb t.llirllichr when the mesh is refined. This intrillsic. ~tlosllticpendcllce is of course lilrlitcd for real ror-its o r 1 t , l ~ lower end by the size of the gmirls. We arltllyzccl the scale invari:~llce of t , l l t a ol,sc$rvcci s11~c2r1)an~lsby cllailgirlg t,ho illcasll siac~or. chcl~iivalenlly t.11~resolution of the systlt:lrl L. 'I'llis t,c,c:hnicpe is equivalenl t o tllc c1assic;~l l,ox-c,olilltilig r r ~ c t l ~ o d .For ' ~ each grid clcrrlcnt i wc: cousitlrr ez , the second invariant of the strain r l ~ t tcnsor ,~ and define the measure pi = ei/ e,. Ld, us dcfine their moments as Mq = py. T h e nornializetl 17 moments mq = ( ~ , / ~ ~ ) ~ ~ % with c c a thc l e system xi xi Fractal Plastic Shear Bands 20 i1 1 577 - ox, averaged over all elements -. .- - oxy averaged over all elements - - - - pressure i n o n e element T i n o n e element TIME Fig. 9 T i m e evolution of a,, and a,, components averaged over the whole grid and evolution of the pressure and the second invariant of the deviatoric stress in one specific element (central one). size. T h e various moments have different scaling exponents d,, defined as rnq oc LWdq. In Fig. 11 we see how d, varies with q for different values of 23. We conclude that within our numerical accur a c y t h e distribution of local shear rate seems to be multifract al. The geometrical fract a1 dimension Do can be exO L - ~ ' . Thus the scalpressed as Mo = L ~ = i n g exponent of the first normalized moment ml t a k e n w i t h positive sign should be compared with t h e geological data on faults. The geometrical Gact a l dimension of the shear band network in our numerical experiment is Do = 1.6f0.1 for B = 1to 10. T o verify this result, we performed a box-counting analysis of a single picture from Fig. 10 with the highest resolution (500 x 500). The zeroth order m o m e n t (or geometrical fractal dimension) gave the same fractal dimension as obtained from different meshes. However, higher moments have different exponents for these two methods. This shows that finite size effects are important and that within t h e range of the sizes that we can simulate, it is n o t possible t o find reliable values for d,. It is even possible that the multifractal nature disappears i n the thermodynamic limit. For smaller B the network of shear b a n d s becomes dense while for B of the order 100 t h e finite size L of the box is felt and so fewer s h e a r bands fit into the system so that the fractal dimension actually decreases. This behavior is d u e to the fact that as seen in Fig. 7, the characteristic length of the problem, namely A S between shear bands increases with B. We therefore propose a two 578 A . N. B. Poliakov et al. ., *: d . Fig. 10 Dependence of the strain rate on different mesh sizes for B = 1 and periodic boundary conditions. Color coding as in Fig. 2(a). F ~ a c t a lPlastic Shear Bands 579 Fig. 11 Dependence of the rnultifractal scaling exponent dg on q for different values B = 0.1,1, 10,100. parameter scaling law of the form: (6) where Mo is the number of plastic elements and F is a scaling function. If we combine the latter expression with a power-law increase of AS cc BY with y = 0.2, we obtain: M~ N L - ~ ~ F ( L B - Y ) We actually plotted our data according t o Eq. (7) and considering that 3 ( x ) xd-dl for small B we achieved reasonable data collapse for dl smaller than 1.6 and y larger than 0.2. These discrepancies are apparently due to the rather small range of system sizes that we can consider. We scanned our sample of Fig. 1 and analyzed the digitalized image by special software (UFC). We obtained a fractal dimension of Do = 1.58 f:0.08. For similar problems, different fractal dimensions have been observed in nature depending on rock type and tectonic evolution. ~ a r t o n analyzed l~ 17 fracture maps at different scales and found dimensions ranging from 1.32 to 1.7 (box-flex method); the analysis of fault systems in iIapanlg yielded values between 1.06 and 1.6 (box-counting technique) and sand box experiments2' gave 1.75 f0.1 (capacity method). One should also note that faults o n the surface of the earth can have different origins. Our numerical result Do = 1.6 zt 0.1 is consistent with some of these field observations and experiments. I n fact, since real faults are three-dimensional, a deviation of our numbers from the fields values would not be surprising. We also analyzed the histogram of the strain rate e by plotting the distribution p ( e ) against e in a log-log plot dividing both axis through log(L). As in the electric analog model for perfect plasticity,21 580 A. N. B. Poliakov et al. Length of shear bond Fig. 12 Length distribution of shear bands for a system of size 500 x 500 for the same physical parameters as in Fig. 2. The straight line is a guide t o t h e eye of slope 2.07. we find that the data collapse for different sizes L on a single hat-shaped curve. This shows the resemblance between fiacture and shear bands. I t is, however, important to point out a crucial difference between the two analyses. In our present analysis, we consider the strain rates, while in most fracture models, the stress is analyzed. We also looked at the length distribution of shear bands. Numerically the length of the bands was obtained using some burning algorithm.22That means starting with one end of a shear band, we iteratively occupy all connected neighboring elements belonging to that shear band. The number of iterations gives the length of the shear band. In Fig. 1 2 we see a log-log plot of this distribution. We do not consider lengths smaller than the width of the bands, i.e., less than 20 lattice units. Beyond that we see a power-law decay with an exponent of r = 2 . l k 0.1. This value is also in agreement with length distributions measured in the field and in experiments. 20923-26 5 . Summary Summarizing, we have found that the plastic behavior of rocks under high confining pressure gives rise to the localization of deformation along a network of shear band with a complex geometry. In spite of the numerous parameters of the model, we have shown that the ratio B of confining pressure to the elastic shear modulus multiplied by the ratio of sound velocity to velocity of loading, represents a charao teristic nondimensional parameter that controls the behavior of the system together with the system size L. We studied the behavior of the system depending on B and L. A characteristic length scale, namely the typical spacing between shear bands in a finite system varies with B. Moreover, the network of shear bands is fractal, and their length distribution follows a power-law. We formulated a two parameter scaling law for the fraction of plastic regions as functions of L and B. The shear rate field forms a "multifractal" measure. Fractal Plastic Shear Bands 581 We also analyzed the fractal dimension of rocks that were sheared under high confining pressure i n t h e past and found a n agreement with our numerical data. Our results are also consistent with previous field observations of fault networks and laboratory experiments. Acknowledgments We t h a n k Ethan Dawson, Peter Cundall and J. P. Villote for useful discussions and suggestions. Raymond Munier generously helped us with analyzing Fig. 1 and calculating fractal dimensions by t h e box-counting technique. Dr. Joe Hull pointed out t o us t h e interesting and beautiful geological region in Pyrenees with t h e shear bands from Fig. 1. This work was completed with t h e support of t h e Swedish N F R grant G-GU 10517-301 during the stay of A. Poliakov i n t h e Laboratoire de Physique e t MBcanique des Milieux HBt6rogBnes (ESPCI). REFERENCES 1. T. von KArrnhn, 2. Ver. Dt. Ing. 55, 1749 (1911). 2. T. F. J. and L. Weiss, Structural Analysis of Metamorphic Tectonites (McGraw-Hill, N.Y., 1963). 3. M. Paterson, Experimental Rock Deformation. The Brittle Field (Springer-Verlag, New York, 1978). 4. J. Desrues, L a localization de la dkformation dans les matkriaux granulaires, PhD thesis, Inst. National Polytech. de Grenoble, 1984. 5. G. Mitra, Geol. Soc. Amer. Bull. 90, 935 (1979). 6. J. Ramsay, J. Struct. Geol. 2 , 83 (1980). 7. B. Velde, et al., J. Geophys. Res. 98, 1193 (1993). 8. P. Vermeer and R. de Borst, Heron 29, 1 (1984). 9. C. H. Scholz, The Mechanics of Earthquakes and Faulting (Cambridge Univ. Press, U.K., 1990). 10. 3. Rudnicki and R. J. R., J. Mech. Phys. Solids 23, 371 (1975). 11. P. Vermeer, Ge'otechnique 40, 223 (1990). 12. P. Cundall, Ingenieur Archiv. 58,148 (1989). 13. P. Cundall, Shear band initiation and evolution in frictional materials. In, Proceedings ASCE Engt neering Mechanics Speciality Conference, Columbus Ohio, 1991. 14. I. Vardoulakis, Int. J. Numer. Anal. Meth. Geomech. 4, 103 (1980). 15. P. Cundall, Numerical modelling of jointed and faulted rock. In, Mechanics of Jointed and Faulted rocks, ed. A. Rossmaaith (A. A. Balkerna, 1990), pp. 11-18. 16. B. Mandelbrot, The Fractal Geometry of Nature (W. H. Freeman and Co., San Francisco, 1982). 17. L. de Arcangelis, In, Statistical Models for the Fracture of Disordered Media eds. H. Herrmann and S. Roux (North Holland, 1990)) p. 229. 18. C. Barton, In, Fractals and their use in the Earth Sciences, eds. C. C. Barton and P. R. LaPointe (to be published in Memoire XX, Geol. Soc. Amer.). 19. T. Hirata, Pageoph 131, 157 (1989). 20. A. Sornette, P. Davy and D. Sornette, J. Geophys. Res. 98, 12111 (1993). 21. S. Roux and A. Hansen, J. Phys. 112, 1007 (1992). 22. H. Herrmann, D. Hong and H. Stanley, J. Phys. A17, L261 (1984). 23. C. Scholz and P. Cowie, Nature 346, 837 (1990). 24. T. Villemin and C. Sunwoo, C.R. Acad. Sci. 305, 1309 (1987). 25. P. Davy, J. Geophys. Res. 98, 12.141 (1993). 26. G. Ouillon, C. Castaing and D. Sornette, Hierarchical Scaling of Faulting (to be published in Europhys. Lett. 1994). 4.1 Stress and strain rate distributions a) b) c) Figure 2: Snapshots of the a) strain rate b) deviatoric stresses c) local rotations for bi-periodic boundary conditions, using = 40 , = 0 , c = 0, = (Poisson ratio, = 0:25), p= = 10;9 or B = p= Vsound =Vbc = 1 Figure 2 shows a snapshot of second invariants of a) strain rate b) deviatoric stress and c) local rotations. We visualize in colors the local changes of the second invariant of the strain rate (Fig. 2a) such that in the yellow and red regions there are strong changes either in the direction or in the magnitude of the motion of the material. The blue regions are elastic. In Fig. 2b high deviatoric stresses are presented by red color and low ones by blue color. In Fig. 2c local rotations are coded such that clockwise rotations are red, anticlockwise ones are blue and green denotes essentially rotationless regions. In all cases the color scale has been chosen to match exactly the range of values obtained in the picture, so that small inhomogeneities may be exaggerated. We see that spontaneously shear bands are formed in which the plastic deformation occurs. These bands form an angle with 45 ; =2 < < (45 ; =2) with respect to the horizontal which is consistent with bifurcation theory?, ?]. a) b) Figure 3: Comparison of a numerical result (accumulated plastic strain) a) with a natural rock from the Pyrenees (b). The scale of (b) is about 2 meters (A 5FF coin can be seen slightly above the center of (a)). In Fig.3 we compare the network of shear bands of natural granite to the results of a simulation with the same parameters as in Fig.2 except that the pressure p is increased by 100 (B = 100). If we identify one discrete element of our simulation with the size of a grain of the rock ( 5mm) the samples of Fig.3a and 3b have about the same size (the number of grains in the sample is the same order as the number of elements in our numerical model). A diculty in comparing the two pictures arises from the fact that we do not know the geological history of the rock so that on one hand we cannot compare the time scales and that on the other hand the natural rock is a cumulation of all shear bands that have occured and one cannot exclude that during the geological deformation the external load changed direction or magnitude. Note that for comparison we do not show the snapshot of strain rate as in Fig. 2a but the accumulated plastic strain after 104 time steps. 4.2 Dependence on parameters First let us discuss the inuence of the dilation and friction angles on the localization. Fig. 4 shows the strain rate for dierent and and identical other parameters. Fig. 4a and Fig. 4d demonstrate that a small dierence of these angles (20 ) gives thick and regular shear bands. An increase of the dierence between these angles enhances the localization and makes shear zones thinner and more irregular (see Fig. 4b with ; = 40 and Fig. 4c with ; = 60 ). It is interesting to note from comparing Figs. 4a and 4d that ; is not sucient to determine the shape of the shear bands but that the absolute value of the angles is important. We performed many numerical experiments studying all possible parameters which can inuence the evolution of the shear bands. The parameters of our model are: 3 = 60o = 0o a) = 40o = 20o c) b) d) Figure 4: Strain rates maps obtained for dierent friction and dilation angles. Note that localization is stronger with increasing of non-associativity ; . Boundary conditions are FBV, system size is 300300 and the other parameters are the same as in Fig. 2. Color coding as in Fig. 2a. - the elastic properties and (since we did not vary the Poisson ratio, one of these two parameters characterize the elastic stiness of the material = ) - the plastic properties which are as discussed above , and c. In the rest of this study, c = 0, = 40 and = 0 . - the system size. Since there is no intrinsic length scale in the problem, only the ratio between the system size and the mesh size is relevant. We call L this ratio. - the loading characteristics, which contains two elements. The initial hydrostatic pressure p and the imposed average shear rate. The latter is characterized by the velocity along the boundary Vbc . Written in dimensionless form, three parameters appears: the system size L, the ratio between the hydrostatic pressure and an elastic constant p= which can be interpreted as proportional to the elastic deformation under the initial hydrostatic pressure and nally the ratio of the sound velocity to the boundary velocity c=Vbc . All other parameters are kept constant in the rest of this work. In fact a useful combination of the latter parameters can be introduced. The shear strain after a time t can be written Vbc t=L. Thus the time at which the yield limit will be reached is y = pL tan()=(Vbc ). The time for an elastic wave to travel through the system is ew = L=c. The ratio of these two intrinsic time scales gives a non dimensional number which we dene as B : (5) B = y = Vpc ew bc where we have dropped the tan() termed which will be kept constant. We found numerically that the shear band patterns only depends on B and L. Changing the pressure p, the shear rate Vbc , the sound velocity c or the elastic properties while keeping B unchanged does not aect the results. This is shown in Fig. 5. We see that the evolution of the system depends only on this non-dimensional parameter B by varying conning pressure, elastic modulus and velocity of loading. Fig. 6 shows that the spacing of shear bands is controlled by this parameter. The decrease in shear band separation with increasing B can be qualitatively explain dthe following way: It is known?, ?, ?], that when a band forms it causes a decrease of mean stress inside of the band due to elastic unloading. The pressure outside the band increases inhibiting the formation of another parallel band nearby. This change in pressure propagates through the system at the sound velocity and thus at high B (large c), the band formation will aect a larger scale. On the contrary, at small B the sound velocity is so low compared to the loading rate that the shear bands have to be more densely distributed to release the imposed shear. The dependence on the other properties , p, ... can be analysed in similar terms using the construction of B . In Fig. 7 we show how the average spacing between the shear bands as measured with a ruler from Fig. 6 depends on B . As we will see in the next section the spacing also depends critically on the resolution L (fractal character) so 4 Vbc Vbc 7 P = 0:1 G = 106 Vsound Vbc = 10 5 P = 103 G = 108 Vsound Vbc = 10 Figure 5: Various parameters having the same value of B and L (B = 1) showing that these two parameters are the only two dimensionless control parameters. Color coding as in Fig. 2a. that it is important in Fig. 7 to work at constant system size. We see that the spacing increases | though slowly | with B in agreement with the previous discussion. The quantitative analysis of this variation is to be taken with caution. Indeed, as can be seen from Fig. 6, when B is less than 1, the shear bands form domains in which bands are all parallel. There are almost no crossing of bands. For larger B , on the contrary, the two families of conjugated shear bands (with opposite slopes) are intermingled and do not form domains. This might explain the apparent discontinuity of the band seperation for small B in Fig. 7. For large B , one begins to be sensitive to the nite size of the system, and the eective separation might be signicantly altered by the presence of periodic boundary conditions. Other boundary conditions are even more strongly biased because of the constraint of the imposed velovity closed to the boundary. A denser distribution of bands arises in this case. However, for the practical purpose of using this separation, we can t roughly a power-law of exponent 0.2 through the data. It is possible however that this apparent behavior is far from what it would be in an innite system size. Another interesting question is to monitor the time evolution of shear bands. To study this problem, we make a vertical cut through a plot of strain rates as ones of Fig. 6 and record the evolution of this one-dimensional slice in time. Figure 8 shows a plot where the horizontal axis gives strain-rates along the cut and the vertical axis is time. Dierent time-dependent behaviours are observed depending on the parameter B . For low B there are regular and stable shear bands. For high B shear bands have a strong time-dependent behaviour and a uctuating intensity. The life time is also shorter for decreasing B . Again this emphasizes the fact that the structuration of the shear banding is essentially a dynamical eect of our model. Figure 9 shows the time evolution of the stress. Stress components (xx and xy ) averaged over all are almost. Both values can be calculated analytically for a single element (xx = ;p (1 + sin()) ; c cos() xy = 0) and we see that in this sense the total system has the same global perfect-plastic behaviour as the single element. On the contrary the stress (pressure and maximum shear stress) in a single element is strongly uctuating in time. Because the material is close/or at the yield the pressure and the shear stress uctuation of a single element are autocorrelated to each other. Thus despite the overall response of the sample is simple and similar to the characteristic of a single element, there is a strong stress and strain rate heteroginity in space and strong uctuations in time for every single element. 4.3 Fractal and multifractal analysis In Fig. 10 we see snapshots from the evolution of the system with dierent mesh sizes L but the same physical parameters. Evidently, because of the underlying instability the shear bands become thinner when the mesh is rened. 5 B=1 B = 100 a) B = 10 c) B = 10000 e) b) d) f) Figure 6: Dependence of the strain rate on the parameter B = p= Vsound =Vbc for systems of size 300300 and periodic boundary conditions. The other parameters are the same as in the Fig 2 and the color coding is like in Fig. 2a. This intrinsic mesh dependence is of course limited for real rocks on the lower end by the size of the grains. We analyzed the scale invariance of the observed shear bands by changing the mesh size or equivalently the resolution of the system L. This technique is equivalent to the classical box-counting method ?]. For grid element i we Pieach consider ei , the second invariant of the strain rate tensor and dene the measure p = e = e . Let us dene their i i i P moments as Mq = i pqi . The normalized moments?] mq = (Mq =M0 )1=q scale with the system size. The various moments have dierent scaling exponents dq , dened as mq / L;dq . In Fig. 11 we see how the dq vary with q for dierent values of B . We conclude that within our numerical accuracy the distribution of local shear rate seems to be multifractal. The geometrical fractal dimension D0 can be expressed as M0 = LD0 = L;d1 . Thus the scaling exponent of the rst normalized moment m1 taken with positive sign should be compared to the geological data on faults. The geometrical fractal dimension of the shear band network in our numerical experiment is D0 = 1:6 0:1 for B = 1 to 10. To verify this result, we performed a box-counting analysis of a single picture from Figure 10 with the highest resolution (500500). The zeroth order moment (or geometrical fractal dimension) gave the same fractal dimension as obtained from dierent meshes. However, higher moments have dierent exponents for these two methods. This shows that nite size eects are important and within range of sizes that we can simulate, it is not possible to nd reliable values for dq and it is even possible that the multifractal nature disappears in the thermodynamic limit. For smaller B the network of shear bands becomes dense while for B of the order 100 the nite size L of the box is felt and so fewer shear bands t into the system so that the fractal dimension actually decreases. This behaviour is due to the fact that as seen in Fig. 7 the characteristic length of the problem, namely S between shear bands increases with B . We therefore propose a two parameter scaling law of the form M0 / L;d1 F (L=S (B )) 6 (6) B = 10. a) B = 100. c) b) d) Figure 8: Time evolution of the shear bands for dierent B. Time increases vertically upwards. Color coding like in Fig. 2a. We also analyzed the fractal dimension of rocks that were sheared under high conning pressure in the past and found an agreement with our numerical data. Our results are also consistent with previous eld observations of fault networks and laboratory experiments. 6 Acknowledgements We thank Ethan Dawson, Peter Cundall and J.P. Villote for useful discussions and suggestions. Raymond Munier generously helped us analysing Fig 1. and calculating fractal dimensions by the box-counting technique. Dr. Joe Hull pointed out to us the interesting and beautiful geological region in Pyrenees with the shear bands from Fig.1. This work was completed with the support of the Swedish NFR grant G-GU 10517-301 during the stay of A. Poliakov in the Laboratoire de Physique et Mecanique des Milieux Heterog"enes, (ESPCI). References 1] 2] 3] 4] 5] 6] 7] 8] 9] 10] 11] T. von Karman, Z. Ver. Dt. Ing. 55, 1749 (1911). T. F.J. and L. Weiss, Structural analysis of metamorphic tectonites, McGraw-Hill, N.Y., 1963. M. Paterson, Experimental Rock Deformation. The Brittle Field, Springer-Verlag, New York, 1978. J. Desrues, La localization de la deformation dans les materiaux granulaires, PhD thesis, Inst. National Polytech. de Grenoble, 1984. G. Mitra, Geol. Soc. Amer. Bull. 90, 935 (1979). J. Ramsay, J. Struct. Geol. 2, 83 (1980). B. Velde, D. Moore, A. Badri, and B. Ledesert, J. Geophys. Res. 98, 1193 (1993). P. Vermeer and R. de Borst, Heron 29, 1 (1984). C. H. Scholz, The mechanics of earthquakes and faulting, Cambridge Univ. Press, U.K., 1990. J. Rudnicki and R. J.R., J.Mech.Phys. Solids 23, 371 (1975). P. Vermeer, Geotechnique 40, 223 (1990). 8 L=100 L=300 a) L=200 c) L=500 e) b) d) f) Figure 10: Dependence of the strain rate on dierent mesh sizes for B=1 and periodic boundary conditions. Color coding as in Fig. 2a. Figure 11: Dependence of the multifractal scaling exponent dq on q for dierent values B = 0:1 1 10 100. 10