JOURNAL OF GEOPHYSICAL RESEARCH, VOL. ???, NO.

, PAGES 1{8,

Brittle-ductile transition and associated seismicity:

Experimental and numerical studies and relationship
with the b-value

David Amitrano
LAEGO, Ecole Nationale Superieure des Mines de Nancy, France

Abstract. The acoustic emission (AE) and the mechanical behavior of granite sam-
ples during triaxial compression tests have been analyzed. The size of AE events displays
power-law distributions, conforming to the Gutenberg-Richter law observed for earth-
quakes which is characterized by the b-value. As the con ning pressure increases, the
macroscopic behavior becomes more ductile. For all di erent stages of the rock mechan-
ical behavior (linear, non-linear pre-peak, non linear post-peak, shearing), there is a sys-
tematic decrease of the b-value with increasing con ning pressure. A numerical model
based on progressive elastic damage and the nite element method allows simulations
of the main experimental observations on AE and of a wide range of macroscopic be-
haviors from brittleness to ductility. The model reproduces a decrease in the b-value that
appears to be related to the type of macroscopic behavior (brittle-ductile) rather than
to the con ning pressure. Both experimental and numerical results suggest a relation-
ship between the b-value and the brittle-ductile transition. Moreover these results are
consistent with recent earthquake observations and give new insight into the behavior
of the Earth's crust.

1. Introduction Fracturing dynamics during mechanical loading, which

can be studied through AE monitoring, usually displays a
The mechanical loading of rocks involves local inelastic power law distribution of acoustic events size.
processes that produce acoustic wave emissions (AE). Non-
linearity of the macroscopic mechanical behavior results N (> A) = c:A
b
(1)
from these microscopic scale processes. For rocks loaded
at high strain rate and low temperature, microfracturing is Where A is the maximum amplitude of AE events, N (> A)
considered to be the main inelastic process [Kranz, 1983]. is the number of events with maximum amplitude greater
The correlation between AE activity and macroscopic in- than A, and c and b are constants. In a log-log representa-
elastic strain has been established in many experimental [see tion, this distribution appears linear and b is given by the
Lockner, 1993, for a review] and numerical [e.g. Young et al., slope of the line.
2000] studies.
As microfracturing progresses, cooperative interactions of
cracks take place and lead to the coalescence of a macro- ( ) = C b:logA
logN > A (2)
scopic fracture, i.e. to the macrorupture [Costin, 1983;
Kranz, 1983; Reches and Lockner, 1994; Schulson et al., This distribution exhibits remarkable similarity to the
1999]. This behavior has been experimentally observed by Gutenberg-Richter relationship observed for earthquakes
AE source location [Lockner et al., 1991]. [Gutenberg and Richter, 1954].
The macroscopic behavior of rocks ranges from brittleness
to ductility depending on rock type and loading conditions
(i.e. strain rate, con ning pressure and temperature). Many ( ) = a bM
logN > M (3)
de nitions of brittle-ductile behavior based on the type of
macroscopic behavior have been proposed [Jaeger and Cook, Where N (> M ) is the number of earthquakes with a mag-
1979]. The most simple is based on the amount of inelastic nitude larger than M .
deformation before the macrorupture (Figure 1). A purely Assuming that the magnitude is proportional to the log of
brittle material fails without any inelastic strain before the the maximal amplitude of the seismic signal, the b-value
failure. By contrast, a purely ductile material strains with- obtained from the magnitude or the amplitude can be com-
out loss of strength. The failure, if any, occurs after a con- pared [Weiss, 1997]. Rigorously, the amplitude measured
siderable amount of inelastic strain. at a given distance from the source should be corrected for
the attenuation. Nevertheless, theoretical [Weiss, 1997] and
experimental studies [Lockner, 1993] have shown that atten-
Copyright 2002 by the American Geophysical Union. uation has no signi cant e ect on the b-value.
As power laws indicate scale invariance and because of the
Paper number . similarities in the physics of the phenomena (wave propaga-
0148-0227/02/$9.00 tion induced by fast source motion), AE of rock observed in
1
2 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY

the laboratory has been considered as a small-scale model for sound velocity is about 4 800 m/s. The mean uniaxial com-
the seismicity in rock masses (rockbursts) or in the Earth's pressive strength is 160 MPa, Young's modulus is 60 GPa
crust (earthquakes) [Scholz, 1968]. Observations of both and the Poisson's ratio is 0.24. The samples were 40 mm in
earthquakes and AE show variations of the b-value in time diameter and 80 mm in length.
and space domains which are usually explained using frac-
ture mechanics theory and/or the self-organized criticality 2.2. Experimental device
(SOC) concept. Mogi [1962] suggested that the b-value de- A hydraulic press of 3000 kN capacity was used. The
pends on material heterogeneity, a low heterogeneity lead- con ning pressure was applied by means of a triaxial cell.
ing to a low b-value. Scholz [1968] observed that the b- The sti ness of the complete9 loading system (press, piston,
value decreases before the maximum peak stress is achieved sample support) is about 10 N/m. The axial displacement
and argued for a negative correlation between b-value and of the platens was measured by an LVDT sensor. The sam-
stress. Main et al. [1989] observed the same variation but ple strain was estimated from the displacement, taking into
invoked a negative correlation between the b-value and the account the sti ness of the loading system (shortening of
stress intensity factor K . Following this idea, Main et al. the piston and the sample support) and the length of the
[1989] proposed di erent patterns of b-value variation be- sample. The axial displacement rate was kept constant near
fore macrorupture, driven by the fracture mechanics and 1 m/s except during the macrorupture when dynamic fail-
the type of rupture (brittle-ductile). The relationship be-
tween the b-value and the fractal dimension D of AE source ure occurs. A resonant transducer (Physical Acoustic Cor-
locations was also investigated [Lockner and Byerlee, 1991] poration,peak frequency : 135 kHz e ective range frequency
and showed a decrease of b-value contemporary to the strain : 100 kHz - 1 MHz) was applied on the outside part of
localization, i.e. to a decrease of D-value. the cell piston which was used as a wave guide. The trans-
Mori and Abercombie [1997] observed a decrease of the ducer was connected to a 40 dB preampli er (PAC 1220A)
b-value with increasing depth for earthquakes in Califor- with adapted lters (20 kHz-1.2 MHz) and then to an AE
nia. They suggested that the b-decrease was related to a analyzer (Dunegan-Endevo 3000 Series) with 40 dB ampli-
diminution of the heterogeneity as depth increases. System- cation which performed the AE counting. In parallel the
atic tests of the dependence of the b-value on depth have signals were digitalized after preampli cation by means of
been recently performed by Gerstenberg et al. [2001] which a fast acquisition board (Imtec T2M50, 8 bits). The sam-
con rm these results. The depth dependence of the b-value pling frequency was 5 MHz and the length of the recorded
have also been observed for the western Alps seismicity [Sue signals was 2048 samples, which corresponds to a duration
et al., 2002] and for earthquakes sequence along the Aswan of 410 s. The signal recording trigger was set to 15 mV
Lake in Egypt [Mekkawi et al., 2002]. and the maximal amplitude to 1 V. The board memory seg-
Other authors have used cellular automata [Chen et al., mentation allowed us to record several hundred signals per
1991; Olami et al., 1992] or lattice solid models [Zapperi second without dead time.
et al., 1997] to simulate power-law distribution of avalanches
which appear to be associated with a ductile macroscopic 2.3. Deformation mode
behavior. Numerical models based on elastic damage [Tang, The Sidobre granite samples were deformed under varying
1997; Tang and Kaiser, 1998] succeed in simulating brittle conditions of con ning pressure, ranging from 0 to 80 MPa.
behavior. Discrete element models simulating macroscopic The axial displacement was applied at a constant strain rate
behavior ranging from brittle to ductile and power-law dis- except during the macrofailure which is unstable. Loading
tributions of earthquakes have also been proposed [Wang was continued after failure until the displacement along the
et al., 2000; Li et al., 2000; Place and Mora, 2000]. Wang
et al. [2000] argue that the b-value depends on the cracks
macrorupture surface reached several millimeters.
density distribution but do not report a relation between
the b-value and the type of mechanical behavior. Amitrano 2.4. Data processing

et al. [1999] proposed a model which simulates both ductile The AE counting was directly obtained from the analyzer.
and brittle behavior and show that the b-value depends on This parameter appeared to be well correlated with the AE
the macroscopic behavior . energy calculated from the digitalized signals. The slope of
These results suggest that a relationship between the b- the cumulative AE counting curve represents the AE activ-
value and the macroscopic behavior may exist. The present ity. The digitalized signals were processed to extract the
paper reports results on AE monitoring of granite samples maximal amplitude and the energy for each signal.
during triaxial compression tests and numerical simulations. The b-value was obtained from the inverse cumulative dis-
We study the e ect of the con ning pressure on both the tribution of the events maximal amplitude. This distribu-
macroscopic behavior and the b-value. tion was tted in a least square sense by a linear function in
a log-log diagram. The slope of this curve gave the b-value.
The error of estimation of the b-value has been calculated for
2. Experimental procedure
a con dence level of 95%. The b-value was rst calculated
2.1. Tested Rock for all recorded events during each test. In order to observe
A set of 34 triaxial compression tests were performed on variations of the b-value during the di erent stages of each
Sidobre granite. This rock contains 71% feldspar, 24.5% test, the b-value was also calculated separately for events oc-
quartz, 4% mica and 0.5% chlorite. The grain sizes are in curring during each stage of the mechanical behavior. The
the range 1-2 mm for the feldspar, 0.5-1 mm for the quartz minimum number of events that was used for calculating the
and 0.5-2 mm for the mica [Isnard, 1982]. The density is 2.65 b-value was xed at 200; according to Pickering et al. [1995]
and the continuity index obtained by sound velocity mea- this population size is acceptable to calculate the b-value
surement (sonic velocity measured on the sample divided by with a good accuracy, i.e. with standard deviation less than
the theoretical value for the intact rock) is about 97%. The 0.1 for the b-value.
 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY 3

3. Experimental results

3.1. Mechanical behavior

el: = E :(1 3 ) (5)
A set of 34 tests have been performed with con ning pres-
initial

sure ranging from 0 to 80 MPa. Figure 2 shows typical Figure 5 presents the mean values of the di erential
results obtained for a con ning pressure of 60 MPa. We stress, 1 3 , and the axial strain, 1 , as a function of the
identify four stages in the mechanical behavior, as observed con ning pressure.
 and  are also plotted. Since
on the () curve, and the acoustic emission activity. the starting and ending points of the unstable failure are
in: in:

 Stage 1 is de ned by the rst linear part of the () strongly machine-dependent, we restrict the discussion to
curve. The initial part of this stage is in uenced by the the values measured at the end of the linear stage 1 and at
closing of microcracks as indicated by the increase in the the peak (end of the stage 2).
slope of the stress-strain curve. After that, the mechanical For 3 = 0, which corresponds to an uniaxial compression
behavior is linear and is not a ected by microcracks. The test, the sample failed immediately after stage 1. The inelas-
AE activity is very low and can be attributed to the closure tic behavior range (i.e.
 and  ) is nearly zero; that
corresponds to a purely brittle behavior. As the con ning
in: in:

or shearing of prexisting cracks [Lockner and Byerlee, 1991].

The b-value is maximum. pressure increases, the stress levels for each stage increase at
 Stage 2 begins with the appearance of a non-linear be- di erent rates. In particular the peak stress increases faster
havior. It corresponds to a strain hardening stage as the than the stress at the end of the linear stage, which indi-
strength increases with rock deformation. AE is caused by cates an increase in strain hardening. In the same manner,
cracks propagation that a ects the macroscopic behavior. the amount of inelastic strain before the peak increases with
The AE activity increases drastically by the end of stage 2 the con ning pressure. Hence, the range of inelastic behav-
and the b-value decreases ( gure 6). ior (i.e.
 and  ) increases with increasing con ning
pressure. This indicates that the pre-peak behavior becomes
in: in:

 Stage 3 corresponds to the post-peak behavior preced- progressively more ductile. The range of con ning pressure
ing the macro-rupture. The rocks display strain softening, that we tested does not cover the entire brittle-ductile tran-
as the strength decreases with increasing strain. AE is pro- sition. Nevertheless, the analysis of the brittle-ductile char-
duced by the propagation and coalescence of cracks. AE acteristic which was performed on granite, a rock commonly
activity reaches its maximum value and the b-value is min- considered as brittle, shows that even this material becomes
imum. This stage ends with the macro-failure which is un- increasingly ductile at relatively low con ning pressure (the
stable. As addressed earlier by Wawersick and Fairhurst maximum con ning pressure of 80 MPa corresponds to a
[1970], this instability occurs when the sample strength de- 3.2 km depth for a natural geostatic stress eld). Similar
creases with strain faster than the apparatus unloads. The results were obtained by Brace et al. [1966] on granite for
starting and ending point of the unstable failure are strongly larger con ning pressure range. They observed an increase
machine-dependent and are not relevant to the description in the range of inelastic behavior, before the peak, for sam-
of the rock sample behavior. It is generally assumed that ple triaxially loaded with con ning pressure ranging from 0
the nucleation of a macroscopic discontinuity occurs simul- to 800 MPa. However, as often observed for granite, the ef-
taneously with the unstable failure. fect of the con ning pressure on the behavior remains minor
 Stage 4 corresponds to the macrorupture surface shear- as the loading ever leads to an unstable failure.
ing. AE is caused by the rupture of surface asperities and
by gouge fracturing. The shear strength is nearly constant 3.3. b-value pressure dependence

or slowly decreases and the AE rate slowly decreases . In order to examine the relationship between the b-value
and the con ning pressure, we calculated the b-value for all
3.2. Brittle-ductile transition
AE events detected during each stage of each one of the tests.
For each test we calculated 1 3 and 1 at the end of Figure 6a displays the cumulative distributions of AE am-
each stage. Figures 3 and 4 display these results for all of plitude for a representative test performed at 3 = 60M P a,
the tests. Stress and strain are plotted as functions of the showing separately the events recorded during each mechan-
con ning pressure. ical stage. As classically observed, the b-value is maximal
In order to estimate the brittle-ductile character of the during stage 1, then decreases during stage 2 and reaches its
mechanical behavior, we quanti ed the range of the inelastic minimal value during stage 3.
behavior before macrorupture using two parameters. One is Figure 6b displays the AE amplitude distributions for
representing the inelastic strain,  , and the second the a set of 4 tests at con ning pressures ranging from 0
stress range of stage 2,
 .
 is the di erence be- to 80 MPa. Each distribution includes all of the events
in:

tween the stress at the end of stage 1 and the peak stress recorded during the test. The b-value decreases with in-
in: in:

(end of stage 2). This parameter quanti es the stress ampli- creasing con ning pressure.
tude of the strain hardening stage. Inelastic strain,  , is Figure 7 displays the b-value corresponding to the dif-
obtained by subtracting the elastic strain,  , to the total ferent stages of mechanical behavior for all of the tests, as
in:

strain, , which is currently measured : a function of 3 . Figure 8 presents the mean b-values for
el:

each stage as a function of the con ning pressure (a) and

tot:

of the di erential stress (b). The b-value is negatively cor-

in: =  (4) related with both the con ning pressure and with the dif-
ferential stress. This behavior is observed for each of the
tot el:

The elastic strain is calculated using the elastic modulus stages as well as for the b-values calculated for the entire
estimated in the linear stage (dotted line in gure 2) and test (i.e. without grouping the recorded events according to
the current value of 1 3 . the stages).
4 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY

For each stage, the b-value, appears to decrease linearly where  is the shear stress;  is the normal stress; C is
with increasing di erential stress. The decreasing rate varies the cohesion; and  is the internal friction angle. In or-
from one stage to the other: the highest rate is observed der to simulate material heterogeneity, the value of the co-
for stage 1. Stages 2 and 3 reveal very similar trends and hesion C is randomly drawn from a uniform distribution.
the smallest e ect of the di erential stress. Stage 4 is an
intermediate case. By contrast, the relationship between This feature is necessary to obtain macroscopic behaviours
the b-value and con ning pressure shows approximately the di ering from those of the elements and a power law dis-
same decreasing rate for all of the stages. One may note tribution of the events size. The Mohr-Coulomb criterion
that the di erential stress and the con ning pressure are allows us to consider separately the e ect of cohesion, in-
not independent parameters, and that their relationships ternal friction angle and con ning pressure. The sensitivity
depend on the mechanical stage ( gure 5a). The relation-
ship between the b-value and the con ning pressure appears studies performed by Amitrano et al. [1999] showed that
to be more general, as the same behavior is observed for all both the b-value and the type of mechanical behavior (brit-
the stages of mechanical behavior. Therefore, the con ning tle/ductile) remained unchanged when changing the cohe-
pressure appears as a more relevant parameter than the dif- sion or the con ning pressure. On the contrary, changing
ferential stress for describing variation in b-value. As far as the internal friction angle modi ed drastically both the b-
we know, this e ect of the con ning pressure on the b-value value and the macroscopic mechanical behavior. Hence, the
has never been experimentally demonstrated before. These
observations are consistent with numerical results obtained key parameter in this model is the internal friction angle, ,
by Amitrano et al. [1999]. which allows us to switch from ductility (  20Æ ) to brittle-
The next section presents further numerical simulations ness (  40Æ ). Figure 9 shows simulation results for tan
of these laboratory experiments. ranging from 0 to 1 ( varying from 0 to 45Æ ). The b-value
for these simulations ranges from 0.63 to 1.2 as the simu-
4. Numerical simulation
lated behavior varies from ductile to brittle. The internal
The model proposed by Amitrano et al. [1999] simulates a friction angle also in uences the nal damage distribution
wide range of mechanical behaviors from ductility to brittle- which varies from di use to localized when  increases. Note
ness, and shows that the simulated b-value and the macro- that in all cases, the damage events are di use in all the sam-
scopic behavior of the model are related. For the present ple at the beginning of the simulation. For the brittle cases,
work, we use this model for simulating both the mechanical
behavior and acoustic emission observed during laboratory the damage localization occurs suddenly during the macro-
tests, and particularly the relationship between the b-value failure. For the intermediate cases, the localization occurs
and the brittle-ductile transition induced by the increase in progressively during the post-peak. For the ductile case the
con ning pressure. damage distribution stays di use during all the simulation.
4.1. Numerical model features The Mohr-Coulomb criterion does not re ect the fact that
The model proposed by Amitrano et al. [1999] is based the internal friction angle decreases as the con ning pres-
on a progressive isotropic damage that is represented by the sure increases, which is usually observed during laboratory
reduction of the elastic modulus. The e ective elastic mod- experiments [see Savage et al., 1996, for a discussion]. A
ulus, E , is expressed as a function of the initial modulus, non-linear criterion is required for taking into account this
, and the damage parameter, D.
ef f:
E ini: pressure dependence of the internal friction angle. Here we
= (1 D):E (6) choose to use the empirical criterion proposed by Hoek and
Brown [1982], which is considered to be relevant for a wide
E ef f: ini:

Such a relationship is valid for a domain which is large as range of rock types.
compared with the defect size. In such a case, the dam-
age can be parameterized by crack density [Kemeny and
Cook, 1986]. The simulated material is discretized using a
r
nite element method with plane-strain hypothesis. The el- 1 = 3 +  c m
3
c
+1 (8)
ement scale is considered as a meso-scale, i.e. intermediate
between the micro-scale corresponding to defects and the where  is the uniaxial compressive strength and m is
macro-scale corresponding to the whole model. The loading
c

consists of increasing the vertical displacement of the upper an empirical parameter controling the pressure dependence
model boundary. After each loading step, when the stress of the criterion slope. A low value of m corresponds to a
of an element exceeds a given strength threshold for dam- faster decrease of the slope with increased con ning pres-
age, its elastic modulus is multiplied by a factor (1 D), sure. The m parameter appears to be empirically related to
D being a constant. Because of the elastic interaction, the the brittle/ductile behavior. A high m value corresponds to
stress redistribution around a damaged element can induce a highly brittle behavior (see Hoek and Brown [1982] for a
damage on adjacent elements and lead to an avalanche of further discussion on the signi cance of this parameter).
damaged elements. The avalanche size corresponds to the
total number of damaged elements during a single loading
step. This size is comparable to the AE activity induced 4.2. Simulation of the Sidobre granite behavior

by stress change [Meredith et al., 1990]. In the original ver- The Hoek and Brown criterion was evaluated for stress
sion of the model the damage threshold was based on the peak values observed in the laboratory ( gure 10), in order
Mohr-Coulomb criterion. to simulate the behavior of the Sidobre granite. The pa-
rameters corresponding to our laboratory results are  = c

 = C + tan (7) 165 MPa and m = 22. These two parameters are used as
 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY 5

input for the numerical simulations. Simulations are per- agreement with the classical observations that distributed
formed for con ning pressure ranging from 0 to 100 MPa. deformation is associated with ductility whereas localized
Typical mechanical and seismic behaviors obtained from deformation is associated to brittleness. In order to quanti-
a simulation performed for 3 =60 MPa are presented in g- tatively estimate the damage localization, we calculated the
ure 11. One may observe that macroscopic non-linear be- spatial correlation dimension of damage, D, using the cor-
havior is associated with the onset of seismic activity. The
seismic activity increases until the macrorupture occurs and relation integral method proposed by Grassberger and Pro-
then decreases during the shearing stage. This behavior dis- coccia [1983]. The spatial correlation integral is de ned as :
plays remarkable similarity with the one observed during
our laboratory tests ( gure 2) and by other authors [e.g.
Brace et al., 1966]. The major di erence with laboratory ( ) = N (N2 1) N (R > r)
C r (9)
tests is that the simulated macrorupture occurs without any
macroscopic strain increase. This is due to the fact that the where N is the total number of damage events, N (R > r)
loading system we simulated has an in nite sti ness. Hence,
there is no elastic energy released by the loading system to is the number of pairs of events separated by a distance
the sample. smaller than r. If this integral exhibits a power-law, C (r) 
Mechanical behaviors obtained from simulations per- r , the population can be considered as fractal and D is
D

formed for di erent values of the con ning pressure are pre- the fractal dimension. In order to estimate the width of the
sented in gure 12. As the con ning pressure increases both power law behavior, we used the two-points slope technique.
the strength and the inelastic range before failure increase. The local curve slope, s , is calculated between every point
This is in agreement with the laboratory observations ( g- l

of the integral (s =
C (r)=
r). The range over which s

ure 5). l l

In order to quantify the brittle-ductile behavior charac- is constant, gives the width of the power-law behavior. For
teristics, values of
 and  (de ned in the preced- this range, C (r) is tted in a least square sense by a linear
ing section) are calculated. These parameters are presented
in: in:

function in a log-log diagram. The slope gives the D-value.

in gure 13. The inelastic range increases with the con- We calculated both b and D values for 75 simulations for
ning pressure in accordance with the laboratory observa- di erent values of  and m parameters. The results are
tions. This indicates that the macroscopic behavior changes c

plotted on the gure 16. The b and D values appear to be

from brittle to more ductile, but without covering the en-
tire brittle-ductile transition. This is due to the fact that negatively correlated. This indicates that di used damage
the simulated material, which is granite, is still brittle at is associated with low b-value, whereas localized damage is
con ning pressures lower than 100 MPa. associated with high b-value.
4.3. Simulation of the Brittle-Ductile transition

This section presents simulations of the behavior of more 5. Discussion

ductile materials for which the complete brittle-ductile tran- 5.1. Brittle-ductile transition as driven by the
sition can be covered within the same range of con ning internal friction angle
pressure as considered above for granite. We simulated the Experimental results have shown that the macroscopic
behavior of materials for which the complete transition is
observable under usual laboratory conditions (i.e. con ning behavior of granite becomes more ductile as the con n-
pressure lesser than 100 MPa): Darley-Dale sandstone and ing pressure increases, even if this change is limited. This
Carrara marble respectively. The parameters of the Hoek change of macroscopic behavior has been quantitatively es-
and Brown criterion corresponding to these materials are timated through the amount of inelastic strain before the
given in table 1. peak. This is a usual result which has been previously ob-
The simulated macroscopic behaviors for these materials served by many authors on di erent rock types [e.g. Jaeger
are presented on gure 14. For both materials, the simu- and Cook, 1979; Hoek and Brown, 1982; Kranz, 1983; Savage
lated macroscopic behavior changes from brittle to ductile
as the con ning pressure increases. This is usually observedet al., 1996; Escartin et al., 1997]. This change of macro-
in laboratory [Jaeger and Cook, 1979; Scholz, 1990; Hoek andscopic behavior is related to a decrease of the internal fric-
Brown, 1982]. tion angle. Because of the correlation between the con ning
Figure 15 presents the b-value as a function of the con- pressure and the internal friction existing for natural ma-
ning pressure for each simulated material. For both sand- terials it is diÆcult to separate experimentally the e ect of
stone and marble, the b-value decreases as the con ning these two parameters on the brittle-ductile transition.
pressure increases for low con ning pressure. It stabilizes to
a value near 1 for higher con ning pressures for which the Using a damage based model, Amitrano et al. [1999] in-
behavior becomes completely ductile (3 > 60 MPa for the vestigated separately the e ect of the con ning pressure,
marble, 3 > 80 MPa for the sandstone). We recall that in the cohesion and the internal friction angle, using the Mohr-
the case of granite, for which the behavior never becomes Coulomb criterion. These results show that the macroscopic
purely ductile, the b-value decreases continuously and does behavior depends only on the internal friction angle and
not reach a minimum. These simulation results suggest that not on the con ning pressure nor on the cohesion. Fig-
the b-value is related to the type of macroscopic behavior ure 9 shows that the type of macroscopic behavior ranges
(brittle/ductile) rather than to the con ning pressure.
from ductility to brittleness for  value ranging from 0 to
4.4. b-value and damage localization 45Æ . This transition is associated with a change on the dam-
As displayed on gure 9, the macroscopic behavior (brit- age localization mode which varies from di use to localized.
tle/ductile) is related to the type of spatial distribution These numerical results are in good agreement with usual
of the damage (localized/di used). This is in remarkable laboratory observations.
6 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY

According to simulations performed with the Mohr- it appears as a relevant parameter for the model we used, as
Coulomb criterion, the behavior does not depend on the this model neglects the detail of low scale micromechanisms
con ning pressure [Amitrano et al., 1999], which is not in and considers the element at meso-scale.
agreement with laboratory observations. This discrepancy One remaining question is how the internal friction acts
can be attributed to the fact that the criterion slope (i.e. in the model to control both the macroscopic behavior, the
the internal friction angle) is not pressure dependent. Using damage localization and the b-value. In a previous study,
a non-linear criterion for which the slope is pressure depen- Amitrano [1999] investigated the in uence of the internal
dent [Hoek and Brown, 1982], we were able to simulate the friction angle on the geometry of interaction between de-
change of mechanical behavior as the con ning pressure in- fects. The authors studied the eld of the Mohr-Coulomb
creases. For this criterion, the m parameter controls the rate criterion (so-called F) around isolated defects (i.e. damaged
of slope decrease, as the con ning pressure increases. This zone). This eld indicates the distance between the stress
parameter has been proposed by Hoek and Brown [1982] to state and the failure criterion. The internal friction angle
be an indicator of the brittleness of the behavior, i.e. a high appeared to strongly in uence the anisotropy of the F eld
m value corresponds to a highly brittle behavior. Simula-
tions performed with di erent values of m corresponding to around the defect. Low friction leads to a quasi-isotropic
di erent rock types (granite, sandstone, marble) show that eld whereas high friction leads to a highly anisotropic eld.
the brittle-ductile transition is obtained at a lower con ning Hence, a low  value allows interaction with other defects
pressure as m is lower. This is in agreement with the in- in any direction (di use interaction). A high value of 
vestigation of Hoek and Brown [1982] on the brittle-ductile induces a strong directionality restricting the possible inter-
transition. action domain (localized interaction). At the macroscopic
Based on these results, we argue that the key factor for scale, this local interaction leads to a damage distribution
the brittle-ductile transition is the internal friction angle which varies form di use to localized, depending on the 
rather than the con ning pressure. value.
However, we stress that the empirical decrease of internal For what concerns the impact on the macroscopic behav-
friction with increased con ning pressure, which is observed ior, we observed that, for brittle behavior, the macrofailure
for all of the rock types we studied (granite, marble, sand- (near instantaneous major stress decrease) of the simulated
stone), re ects physical processes that are fundamentally sample was associated with a sudden localization of the dam-
di erent. For the granite, as for other crystalline rocks, it age. Hence, the elastic energy contained in all the sample is
has been shown that increased con ning pressure leads to a released into a localized area which leads to a huge damage
change in micro-fracturing processes [Escartin et al., 1997; event and consequently to a signi cant stress decrease. On
Velde et al., 1993; Jaeger and Cook, 1979]. At low con n- the other hand, for a ductile behavior, the damage localiza-
ing pressure, the dominant process is tensile cracks (mode I) tion, if any, occurred progressively, in the absence of stress
parallel to the main stress direction. As the con ning pres- decrease. Hence, we suggest that this kind of macroscopic
sure increases, tensile cracks are progressively replaced by behavior results from the damage localization mode, which
shear cracks (mode II-III). This change on the micro-scale is controlled by the internal friction angle.
processes is associated with a decrease of the internal fric- For what concerns the relationship between the friction
tion, as mode II-III cracks are less pressure sensitive than angle and the b-value, we can propose the following ex-
mode I cracks. planation, which is also based on the interaction geometry.
For porous rocks like sandstone, brittle-ductile transition We have seen previously that the isotropy/anisotropy of the
is associated with the transition from dilating shear bands damage criterion (F) determines the geometrical interaction
to compacting shear bands [Menendez et al., 1996; Besuelle, between defects. For an isotropic eld, when a damage event
2001]. As the con ning pressure increases, intergranular occurs into an element, it can induce damage in all the sur-
cracking is progressively replaced by crushing and pore col- rounding elements. This facilitates the emergence of large
lapse. The e ect of con ning pressure is to impede dilatant events, which corresponds to a low b-value. On the other
deformation which is replaced by a non-dilatant one. hand, an anisotropic eld allows interaction preferentially
For rocks constituted by more ductile minerals, at room in the direction of anisotropy. This restricts the emergence
temperature, or when the temperature is high, one must of large events, which corresponds to a high b-value. This
consider the competition between cracking and crystal plas- qualitative explanation is con rmed by the quantitative re-
ticity. On one hand, crack propagation occurs when the sults presented in gure 16, where the damage localization
tensile strength at the crack tips is reached. The increase is estimated by the D-value.
of con ning pressure tends to close the cracks and impedes
their propagation. On the other hand, dislocation glide, in-
volved in mineral plasticity, depends only on the applied 5.2. Brittle-ductile transition, b-value and earth

shear stress, and consequently is relatively insensitive to crust behavior

Experimental results show that the rock behavior be-

the con ning pressure. Hence, as the con ning pressure in-
creases, crack propagation is progressively replaced by plas- comes more ductile as the con ning pressure increases. We
ticity. This has been experimentally observed for Carrara also observe a systematic decrease of the b-value as the con-
marble at room temperature [e.g. Fredrich et al., 1989], as ning pressure increases. This suggests that a relationship
for quartz and olivine at higher temperature [e.g. Darot may exist between the b-value, re ecting the damage dy-
et al., 1985; Hirth and Tullis, 1992b, a]. namics at low scale, and the macroscopic behavior (brittle-
All these observations show the diversity of the micro- ductile). The decrease of b-value with con ning pressure
scale processes involved in the brittle-ductile transition. is observed for di erent types of mechanical behavior and
However, they all show a decrease of the internal friction particularly for shear faulting, which is generally supposed
angle as the con ning pressure increases. The variations of to be the main mechanism for earthquakes [Scholz, 1990].
such a parameter capture a wide range of processes. Hence, Therefore our results can give new insights into earthquakes
 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY 7

dynamics. In our experimental study, the con ning pressure Both experimental and numerical results are consistent
ranges from 0 to 80 MPa, which corresponds roughly to a with earthquake observations, which show a decrease in b-
natural stress state at depths ranging from 0 to 3200 m. value with depth and give a new explanation for this de-
These experimental results suggest that, in this range, the crease. The b-value decrease could result from the change
b-value variation with depth can reach 0.5. from brittleness to ductility as the depth increases.
Numerical simulations show that the b-value depends on
the macroscopic behavior, which is controlled by the inter- Acknowledgments. The author thanks Y.Orengo and D.
nal friction angle rather than by the con ning pressure. De- Hantz for their technical support during the laboratory experi-
pending on the rock type, the decrease of the b-value is ob- ments; JR Grasso, F. Lahaie, A. Helmstetter and J. Weiss for
tained at di erent con ning pressures. The decrease of the helpful comments; A. Rouleau and H. Lopez for improvements of
b-value is obtained at lesser con ning pressure for a ductile the manuscript; G. Hirth, N. Beeler and an anonymous reviewer
whose critical remarks helped to improve this paper. This work
material than for a brittle one. Based on these results, we was supported by the French program INSU-ACI CATNAT.
suggest that the b-value is related to the type of macroscopic
behavior (brittle/ductile), rather than to con ning pressure.
Both experimental and numerical results are in agreement
with several earthquake observations. Mori and Abercombie References
[1997] observed a depth dependence of b-value for earth-
quakes of California between 1974 and 1996. Based on the Amitrano, D., Emission acoustique des roches et endommage-
ment: Approches experimentale et numrique, Application a la
study of Mogi [1962], they invoked the e ect of heterogene- sismicite miniere, Phd thesis, Universite Joseph Fourier, 1999.
ity, which they assumed to be decreasing with depth and Amitrano, D., J.-R. Grasso, and D. Hantz, From di use to local-
could result in a reduced b-value. We suggest that such ized damage through elastic interaction, Geophysical Research
depth dependence of the b-value could also be due to an Letters, 26(14), 2109{2112, 1999.
increase in con ning pressure as depth increases. Our labo- Besuelle, P., Compacting and dilating shear bands in porous rock:
ratory results are comparable to these observations but ar- Theoretical and experimental conditions., Journal of Geophys-
gue for a di erent explanation for the b-value decrease. In ical Research, 106(B7), 13435{13442, 2001.

our experiments, the heterogeneity is constant and the only Brace, W., B. Paulding, and C. Scholz, Dilatancy in the fracture
of cristalline rocks, Journal of Geophysical Research, 71(16),
variable parameter is the con ning pressure. Moreover, the 3939{3953, 1966.
numerical simulations show that the brittle-ductile transi- Chen, K., P. Bak, and S. Obukhov, Self-organized criticality in a
tion is controlled by the decrease in the internal friction crack-propagation model of earthquakes, Physical Review A,
angle, induced by the pressure increase rather than by the 43(2), 625{630, 1991.
con ning pressure itself. We suggest that the b-value de- Costin, L., A microcrack model for the deformation and failure of
crease is related to the change of mechanical behavior with brittle rock, Journal of Geophysical Research, 88(B11), 9485{
depth, rather than to a change of heterogeneity nor to an 9492, 1983.
increase of con ning pressure. Darot, M., Y. Gueguen, and Z. Bencheman, Ductile-brittle tran-
Our results are also in agreement with Gerstenberg et al. sition investigated by micro-indentation : results for quartz
and olivine, Phys. Earth and Plan. Int., 40, 180{186, 1985.
[2001] who performed systematic tests of the dependence Escartin, J., G. Hirth, and B. Evans, Non dilatant brittle defor-
of the b-value on depth at California. They observed that, mation of serpentinites: Implications for Mohr-Coulomb the-
for more than 70% of the studied area, the b-value is sig- ory and the strength of faults, Journal of Geophysical Re-
ni cantly smaller in depth. According to these authors, the search, 102(B2), 2897{2913, 1997.
areas where depth dependence of the b-value is not observed Fredrich, J., B. Evans, and T. Wong, Micromechanics of the brit-
correspond to zones where the geostress state shows anoma- tle to plastic transition in Carrara marble, Journal of Geo-
lous variation with depth. Our results suggest that this physical Research, 4129-4145, 1989.

could be related to a particular trend of the con ning pres- Gerstenberg, M., S. Wiemer, and D. Giardini, A systematic test
of the hypothesis that b varies with depth in California, Geo-
sure with depth or to regional variations of the mechanical physical Research Letters, 28(1), 57{60, 2001.
behavior of the crust, which can be more or less brittle or Grassberger, P., and I. Prococcia, Measuring the strangeness of
ductile. Recently, a statistical analysis of the western Alps stange attractors, Physica, 9, 189{208, 1983.
seismicity between 1987-1999 highlighted signi cant di er- Gutenberg, B., and C. Richter, Seismicity of the earth and asso-
ence in b-value observed for two seismic zones characterized ciated phenomena. Princeton University Press, Princeton, 2nd
by di erent earthquake depth distributions [Sue et al., 2002]. edition edn., 1954.
The deeper one displays a lower b-value. Similar observa- Hirth, G., and J. Tullis, The brittle-plastic transition in experi-
tions have been done for the 1982-1999 earthquakes sequence mentally deformed quartz aggregates, Journal of Geophysical
along the Aswan Lake in Egypt [Mekkawi et al., 2002]. Research, 99(B6), 11731{11747, 1992a.
Hirth, G., and J. Tullis, Dislocation creep regimes in quartz aggre-
gates, Journal of Structural Geology, 14(2), 145{159, 1992b.
6. Conclusion Hoek, E., and E. Brown, Underground excavations in rock. Insti-
tution of Mining and Metallurgy, London, 1982.
Experimental results shows two e ects of the increase of Isnard, P., Etude chimique du massif granitique du Sidobre
con ning pressure on the behavior of granite samples. On (Tarn), Technical report, Ecole National de Geologie, 1982.
one hand the behavior becomes more ductile as the con n- Jaeger, J. C., and N. G. W. Cook, Fundamentals of rock mechan-
ing pressure increases. On the other hand, the b-value of ics. Chapman and Hall, London, 1979.

acoustic emission size associated with di erent mechanical Kemeny, J., and N. G. W. Cook, E ective moduli, Non-linear de-
behavior decrease with the con ning pressure. The numeri- formation and strength of a cracked elastic solid, Int. J. Rock
Mech. Min. Sci. and Geomech. Abstr., 23(2), 107{118, 1986.
cal model we proposed is able to simulate the brittle-ductile Kranz, R., Microcracks in rocks: a review, Tectonophysics, 100,
transition as the con ning pressure increases. Numerical re- 449{480, 1983.
sults show that the b-value is controlled by variations of the Li, H., Y. Bai, M. Xia, M. Ke, and X. Yin, Damage localisation
internal friction angle induced by changes on the con ning as a possible precursor of earthquake rupture, Pageoph, 157,
pressure. 1929{1943, 2000.
8 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY

Lockner, D., The role of acoustic emission in the study of rock Scholz, H., The mechanics of earthquakes and faulting. Cam-
fracture, Int. J. Rock Mech. Min. Sci. and Geomech. Abstr., bridge University Press, 1990.
30(7), 883{899, 1993. Schulson, E., D. Illiescu, and C. Renshaw, On the initiation of
Lockner, D., and J. Byerlee, Precursory AE patterns leading to shear faults during brittle compressive failure: A new mech-
rock fracture, in Vth Conf. AE/MS Geol. Str. and Mat., edited anism, Journal of Geophysical Research, 104(B1), 695{705,
by Hardy, pp. 45{58, The pennsylvania State University. Trans 1999.
Tech Publication, Clausthal-Zellerfeld, Germany, 1991.
Lockner, D., J. Byerlee, V. Kuskenko, A. Ponomarev, and Sue, C., J. Grasso, F. Lahaie, and D. Amitrano, Mechanical be-
A. Sidorin, Quasi-static fault growth and shear fracture en- havior of western alpine structures inferred from statistical
ergy in granite, Nature, 350, 39{42, 1991. analysis of seismicity, Geophysical Research Letters, 29(8), 65{
Main, I., P. Meredith, and C. Jones, A reinterpretation of the 1 65{4, 2002.
precursory seismic b-value anomaly from fracture mechanics, Tang, C., Numerical simulation of progressive rock failure and
Geophysical Journal, 96, 131{138, 1989. associated seismicity, Int. J. Rock Mech. Min. Sci. and Ge-
Mekkawi, M., A. Hassoup, J. Grasso, and P. Schnegg, Fractal and omech. Abstr., 34(2), 249{261, 1997.
spectral analysis of the earthquake sequences along the Aswan Tang, C., and P. Kaiser, Numerical simulation of cumulative dam-
Lake, Egypt, in 26th General Assembly of the European Geo- age and seismic energy release during brittle rock failure - Part
physical Society, vol. Geophysical Research Abstracts, Volume
4, 2002, pp. abstract nEGS02{A{00027, Nice. 2002. I: Fundamentals, Int. J. Rock Mech. Min. Sci. and Geomech.
Abstr., 35(2), 113{121, 1998.
Menendez, B., W. Zhu, and T.-F. Wong, Micromechanics of brit-
tle faulting and cataclastic ow in Berea Sandstone, Journal Velde, B., D. Moore, A. Badri, and B. Ledesert, Fractal analy-
of Structural Geology, 18(1), 1{16, 1996. sis of fractures during brittle to ductile changes, Journal of
Meredith, P., I. Main, and C. Jones, Temporal variation in Geophysical Research, 98(B7), 11935{11940, 1993.
seismicity during quasi-static and dynamic failure, Tectono- Wang, Y., X. Yin, M. Ke, M. Xia, and K. Peng, Numerical sim-
physics, 175, 249{268, 1990. ulation of rock failure and earthquake process on mesoscopic
Mogi, K., Magnitude frequency relations for elastic shocks ac- scale, Pageoph, 157, 1905{1928, 2000.
companying fractures of various materials and some related Wawersick, W., and C. Fairhurst, A study of brittle rock fracture
problems in earthquakes, Bull. Earthquake Res. Inst. Univ. in laboratory compression experiments, Int. J. Rock Mech.
Tokyo, 40, 831{853, 1962.
Min. Sci. and Geomech. Abstr., 7, 561{575, 1970.
Mori, J., and R. Abercombie, Depth dependence of earth-
quake frequency-magnitude distributions in California: Impli- Weiss, J., The role of attenuation on acoustic emission ampli-
cation for rupture initiation, Journal of Geophysical Research, tude distributions and b-values., Bulletin of the Seismological
102(B7), 15081{15090, 1997. Society of America, 87(5), 1362{1367, ns, 1997.
Olami, Z., H. Feder, and K. Christensen, Self-organised criticality Young, R., J. Hazzard, and W. Pettitt, Seismic and microme-
in a continuous, nonconservative cellular automaton modeling chanical studies of rock fracture, Geophysical Research Letters,
earthquake., Physical Review Letters, 68(8), 1244{1247, 1992. 27(12), 1767{1770, 2000.
Pickering, G., J. Bull, and D. Sanderson, Sampling power-law Zapperi, S., A. Vespignani, and E. Stanley, Plasticity and
distributions, Tectonophysics, 248, 1{20, 1995.
Place, D., and P. Mora, Numerical simulation of Localisation phe- avalanche behaviour in microfracturing phenomena, Nature,
388(14 august 1997), 658{660, 1997.
nomena in a fault zone, Pageoph, 157, 1821{1845, 2000.
Reches, Z., and D. Lockner, Nucleation and growth of faults in
brittle rocks, Journal of Geophysical Research, 99(B9), 18159{ D. Amitrano, Laboratoire Environnement Geomecanique et
18173, 1994. Ouvrages, Ecole Nationale Superieure des Mines de Nancy, Parc
Savage, J., J. Byerlee, and D. Lockner, Is internal friction friction, de Saurupt, 54042 Nancy Cedex, France.
Geophysical Research Letters, 23(5), 487{490, 1996.
Scholz, C. H., The frequency-magnitude relation of microfractur-
ing in rock and its relation to earthquakes, Bulletin of the (Received XX; revised XX;
Seismological Society of America, 58(1), 399{415, 1968. accepted XX.)
 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY 9

Brittleness Ductility
σ σ

rupture rupture

inelastic strain
ε ε
Figure 1. Brittleness and ductility as characterized by the
stress-strain curve (). Brittleness is characterized by the
absence of inelastic strain before failure. On the contrary
ductility involves a considerable inelastic strain before fail-
ure.
σ3=60 MPa x 10
4

600 15

1 2 3 4
AE
500

400 10

Σ AE Energy
σ1- σ3 (MPa)

rupture
Macrorupture

300

σ -σ
1 3
Unstable

200 5

100

0 0
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035
ε1

Figure 2. Typical mechanical behavior observed for tri-

axial compression tests. 1 is the longitudinal compressive
stress, 3 the con ning pressure, 1 the longitudinal strain.
The di erential stress, 1 3 , and the cumulative acoustic
emission energy, AE , are plotted as functions of 1 .
Stage 1 corresponds to the initial linear behavior, stage 2, to
Energy

the non-linear pre-peak behavior, stage 3, to the post-peak

behavior which leads to the macro-rupture (shaded area)
and stage 4, to the shearing along the macro-rupture sur-
face.
Type of c m
Material MPa
Sidobre granite 165 22
Darley Dale sandstone 70 10
Carrara marble 100 10
Table 1. Parameters of the Hoek and Brown criterion used for the brittle-ductile simulations.
10 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY

700
Linear (a) Peak (b)
400 600
mean mean

500
300
σ −σ (MPa)

σ1−σ3 (MPa)

400
3

200 300
1

200
100
100

0 0
0 20 40 60 80 0 20 40 60 80
σ3 (MPa) σ3 (MPa)
700 350
Postpeak (c) ∆σ (d)
in.
600 mean 300
mean
500 250
σ −σ (MPa)

∆ σin. (MPa)

400 200
3

300 150
1

200 100

100 50

0 0
0 20 40 60 80 0 20 40 60 80
σ (MPa) σ (MPa)
3 3

Figure 3. Stress level at the end of the linear stage (a),

at the stress peak (b), at the end of the post peak (c), and

 (d). 1 3 is plotted as a function of the con n-
ing pressure. Open circles correspond to individual values.
in:

Black squares correspond to the mean values calculated for

each value of con ning pressure. Vertical error bars indicate
the standard deviation.
 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY 11

0.015 0.015
Linear (a) Peak (b)
mean mean

0.01 0.01
ε1

ε1
0.005 0.005

0 0
0 20 40 60 80 0 20 40 60 80
σ31 (MPa) σ3 (MPa)
−3
x 10
0.015
Postpeak (c) Inelastic strain at peak (d)
mean mean
4

0.01
ε1

ε1

2
0.005

0 0
0 20 40 60 80 0 20 40 60 80
σ3 (MPa) σ3 (MPa)

Figure 4. Axial strain at the end of the linear stage (a),

at the stress peak (b), at the end of the post-peak stage (c)
and inelastic strain at peak (d). Values are plotted as func-
tions of the con ning pressure. Open circles correspond to
individual values. Black squares correspond the mean val-
ues calculated for each value of con ning pressure. Vertical
error bars indicate the standard deviation.

0.015
Linear Linear
600 Peak
peak
postpeak Postpeak
∆σ ε at peak
500 in. in.

0.01
σ1 − σ3 (MPa)

400
ε1

300

0.005
200

100

0 0
0 20 40 60 80 0 20 40 60 80
σ (MPa) σ (MPa)
(a) (b)
Figure 5. Mean value of the stress (a) and strain (b) for
the di erent stages of the mechanical behavior and

and  as a function of the con ning pressure. Vertical
in:

error bars indicate the standard deviation.

in:
12 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY

4 All events b=1.2± 0.02 4

10 10
Stage 1 b=1.9± 0.08
Stage 2 b=1.2± 0.03
Stage 3 b=1.1± 0.01
10
3 Stage 4 b=1.2± 0.02
3
10
)

)
max

max

2
10
N(>A

N(>A

10
2
0 MPa, b=1.63 ± 0.02
± 0.01
1 20 MPa, b=1.5
10
40 MPa, b=1.24 ± 0.01
60 MPa, b=1.22 ± 0.02
± 0.02
80 MPa, b=1.06
0 1
10 −2 −1 0
10 −2 −1 0
10 10 10 10 10 10
A (mV) A (mV)

(a) (b)
Figure 6. Amplitude distribution of AE event and esti-
mated b-value. a) For the di erent states of mechanical
behavior observed during a single triaxial compression test
performed at 3 =60 MPa. b) For a set of tests performed at
di erent values of con ning pressures, each distribution in-
cluded all the events recorded during each test. The b-values
are given with an estimation error for a 95 % con dence
level.
 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY 13

2 2
(a) All events (b) Linear
1.8 mean 1.8 mean

1.6 1.6
b-value

b-value
1.4 1.4

1.2 1.2

1 1

0.8 0.8
0 20 40 60 80 0 20 40 60 80
σ (MPa) σ (MPa)
3 3
2 2
(c) Nonlinear (d) Shear
1.8 mean 1.8 mean

1.6 1.6
b-value

b-value

1.4 1.4

1.2 1.2

1 1

0.8 0.8
0 20 40 60 80 0 20 40 60 80
σ (MPa) σ (MPa)
3 3

Figure 7. b-value calculated for all AE events recorded

during the entire test (a) or for events recorded during each
di erent stage of the mechanical behavior (b-d). The b-value
is given with an estimation error for a 95 % con dence level.
14 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY

Stage 1
Stage 2
2 2 Stage 3
Stage 4
1.8 1.8
b−value

b−value

1.6 1.6

1.4 1.4

1.2 All events 1.2

Stage 1
1 Stage 2 1
Stage 3
Stage 4
0.8 0.8
0 20 40 60 80 0 100 200 300 400 500 600
σ (MPa) σ −σ (MPa)
(a) (b)
Figure 8. Mean b-value calculated for AE events recorded
during the di erent stages of the mechanical behavior. a) as
a function of the con ning pressure; b) as a function of the
di erencial stress 1 3 . Horizontal and vertical bars give
the standard deviation.
 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY 15

Mechanical behavior
and seismicity Event size distribution Final damage state
100 400
3 ± 0.01
10 b=0.63
σ1 (MPa)

N(>s)
2
tanφ=0 50 200 10 40
1
10
0 20
0 0 10
0 0.005 0.01 0 1 2 3
10 10 10 10
200 400
150 300
3
10 b=0.68± 0.01 50
σ1 (MPa)

N(>s)
tanφ=0.2 100 200 10
1
50 100 10
0
0 0 10 0
0 0.005 0.01 0 1 2 3
10 10 10 10
200 200
3 ± 0.01
150 150 10 b=0.8
σ1 (MPa)

50
N(>s)

2
tanφ=0.4 100 100 10
1
50 50 10
0
0 0 10 0
0 0.005 0.01 0 1 2 3
10 10 10 10
200 2000
150 1500
3
10 b=0.85± 0.01
σ1 (MPa)

N(>s)

2
tanφ=0.6 100 1000 10 50
1
50 500 10
0
0 0 10 0
0 0.005 0.01 0 1 2 3
10 10 10 10
200 4000
3 ± 0.05
100
150 3000 10 b=1.1
σ1 (MPa)

N(>s)

2
tanφ=0.8 100 2000 10 50
1
50 1000 10
0
0 0 10 0
0 0.005 0.01 0 1 2 3
10 10 10 10
200 4000
3 ± 0.07 100
150 3000 10 b=1.2
σ1 (MPa)

N(>s)

2
tanφ=1 100 2000 10 50
1
50 1000 10
0
0 0 10 0
0 0.005 0.01 0 1 2 3
10 10 10 10
ε1
Event size (s)
Figure 9. Simulation results performed using the Mohr-
Coulomb criterion. Parameters: E=50 GPa,  =0.25,
D =0.05, C : randomly drawn between 25 and 50 MPa, tan
ranging from 0 to 1.
16 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY

700
Data
HB criterion
600

500
σ (MPa)

400

300
1

200

100

0
0 20 40 60 80
σ (MPa)
Figure 10. 1 at the peak as a function of the con ning
pressure 3 for all the performed test (open circles) and the
corresponding Hoek and Brown criterion ( =165 MPa and
m=22).
c

500 σ -σ
1 2 1 3

400 Macrorupture
σ1- σ3 (MPa)

300

3
200

100

0
0 0.005 0.01 0.015
ε1
2000 Seismic Event size count
Seismic Event size count

1500

1000

500

0
0 0.005 0.01 0.015
ε1

Figure 11. Results obtained from a simulation performed

at 3 =60 MPa. Di erential stress, 1 3 , and seismic event
size count are plotted as a function of longitudinal strain,
1 . Stage 1 corresponds to the linear behavior without seis-
mic activity. Stage 2 corresponds to the onset of seismicity
and of non-linear behavior which leads to the macrorupture.
Stage 3 corresponds to fault shearing.
 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY 17

600 σ3=100 MPa

σ =80 MPa
500 3

σ3=60 MPa
σ1−σ3 (MPa)

400
σ3=40 MPa
300
σ =20 MPa
3
200

σ3=0
100

0
0 0.005 0.01 0.015
ε
1

Figure 12. Mechanical behavior observed for simulations

of the granite performed at con ning pressure values ranging
from 0 to 100 MPa.

0.014
Linear Linear
600 Failure
failure εin.
∆σ 0.012
in.
500
0.01
σ −σ (MPa)

400
0.008
ε1
3

300
0.006
1

200
0.004

100 0.002

0 0
0 20 40 60 80 100 0 20 40 60 80 100
σ (MPa) σ (MPa)
3 3

(a) (b)
Figure 13. Stress and strain levels obtained from simu-
lations for di erent mechanical behaviors stages and

and
 as a function of the con ning pressure.
in:

in:
18 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY

400 400

σ3=100 MPa
σ3=100 MPa
300 300
σ =80 MPa
σ3=80 MPa 3
σ −σ (MPa)

σ −σ (MPa)
σ =60 MPa
3
200 200
σ3=60 MPa
3

3
1

1
σ3=40 MPa
σ3=40 MPa σ =20 MPa
3
100 100
σ =20 MPa
3
σ3=0 MPa
σ3=0 MPa
0 0
0 0.005 0.01 0.015 0 0.005 0.01 0.015
ε ε
1 1

(a) (b)

Figure 14. Simulation of the brittle-ductile transition for the Darley-Dale sandstone (a) and the Carrare marble (b).

2.6
Granite
2.4 Sandstone
Marble
2.2

2
b−value

1.8

1.6

1.4

1.2

0.8
0 20 40 60 80 100
σ3 (MPa)

Figure 15. b-value as a function of con ning pressure re-

sulting from simulations of the brittle-ductile transition for
granite, sandstone and marble.
 AMITRANO: BRITTLE-DUCTILE TRANSITION AND ASSOCIATED SEISMICITY 19

1.8

1.7 D = − 0.55*b + 2
r = − 0.86
1.6

1.5
D−value

1.4

1.3

1.2

1.1

1
0.6 0.8 1 1.2 1.4 1.6 1.8 2
b−value

Figure 16. D-value as a function of b-value for 75 simula-

tions performed with various values of  and m parameters.
D and b appear to be negatively correlated.
c