Journal of the Mechanics and Physics of Solids

49 (2001) 21132132
www.elsevier.com/locate/jmps
Continuum and atomistic studies of intersonic
crack propagation
Huajian Gao
a,
, Yonggang Huang
b
, Farid F. Abraham
c
a
Division of Mechanics and Computation, Stanford University, Stanford, CA 94305, USA
b
Department of Mechanical and Industrial Engineering, University of Illinois, Urbana, IL 61801, USA
c
IBM Research Division, Almaden Research Center, San Jose, CA 95120-6099, USA
Abstract
Mechanisms of intersonic crack propagation along a weak interface under shear dominated
loading are studied by both molecular dynamics and continuum elastodynamics methods. Part
of the objective is to test if continuum theory can accurately predict the critical time and length
scales observed in molecular dynamics simulations. To facilitate the continuum-atomistic link-
age, the problem is selected such that a block of linearly isotropic, plane-stress elastic solid
consisting of a two-dimensional triangular atomic lattice with pair interatomic potential is loaded
by constant shear velocities along the boundary. A pre-existing notch is introduced to represent
an initial crack which starts to grow at a critical time after the loading process begins. We
observe that the crack quickly accelerates to the Rayleigh wave speed and, after propagating at
this speed for a short time period, nucleates an intersonic daughter crack which jumps to the
longitudinal wave speed. The daughter crack emerges at a distance ahead of the mother crack.
The challenge here is to test if a continuum elastodynamics analysis of the same problem can
correctly predict the length and time scales observed in the molecular dynamics simulations. We
make two assumptions in the continuum analysis. First, the crack initiation is assumed to be
governed by the Grith criterion. Second, the nucleation of the daughter crack is assumed to
be governed by the BurridgeAndrew mechanism of a peak of shear stress ahead of the crack
tip reaching the cohesive strength of the interface. Material properties such as elastic constants,
fracture surface energy and cohesive strength are determined from the interatomic potential. Un-
der these assumptions, it is shown that the predictions based on the continuum analysis agree
remarkably well with the simulation results. ? 2001 Elsevier Science Ltd. All rights reserved.
Keywords: Crack propagation; Continuum; Atomistic

Corresponding author. Tel.: +49-0711-2095-519; fax: +49-0711-2095-520.

E-mail address: hjgao@mf.mpg.de (H. Gao).
0022-5096/01/$ - see front matter ? 2001 Elsevier Science Ltd. All rights reserved.
PII: S0022- 5096( 01) 00032- 1
2114
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
1. Introduction
This paper reports a study of intersonic crack dynamics using both atomistic and
continuum methods, with emphasis on the linkage between mechanics and physics
modeling of fracture.
1
Analyses combining atomistic and continuum methods will be especially useful for
nanoscale modeling. Continuum mechanics is limited by its coarse view of physical
phenomena and various assumptions adopted in its constitutive laws. On the other
hand, atomistic methods, such as molecular dynamics (MD), are limited not just by
the large number of degrees of freedom involved but also by the time scale. This
is usually in the range of picoseconds and hence the modeling of any process that
might take a considerable amount of time (such as diusion) is dicult. As a result of
this gap between continuum and atomistic methods, there are only a few problems to
which continuum and atomistic approaches can both be applied and the results directly
compared. One of such problems is the simulation of dynamic crack propagation in
a nanometer size crystal. The system is large enough for the continuum methods of
dynamic fracture mechanics (Freund, 1990; Broberg, 1999) to be applicable. On the
other hand, the emergence of large scale parallel computers has also allowed MD
simulations of crack propagation for system sizes reaching 1 billion atoms and time
scales approaching nanoseconds (Abraham et al., 1998; Abraham, 2001). Consider the
fact that linear elastic wave speeds are typically of the order of a few kilometers
per second, or a few nanometers per picosecond. These are precisely the length and
time scales suitable for MD. Atomistic simulations provide an ab initio investigation
of fracture by which the validity of continuum methods can be tested. Continuum
mechanics analysis provides a conceptual framework in which MD simulation data can
be analyzed and understood. It is in this spirit that we conduct a joint continuum and
atomistic investigation of intersonic crack propagation along a weak interface.
Our present study is motivated by recent experiments on intersonic crack propagation
by Rosakis et al. (1999), who investigated shear dominated crack growth along the
weak planes in a brittle polyester resin under far eld asymmetrical loading. They
observed crack propagation as fast as the longitudinal wave speed. This experiment is
interesting because it has been widely believed that a brittle crack cannot propagate
faster than the Rayleigh wave speed. The origin of this belief stems from the predictions
of continuum mechanics with regard to the dynamic elastic solutions of the near-tip
stress elds and energy release rates for various velocity regimes and dierent types
of external loading. Broberg (1999) and Freund (1990) have elegantly summarized the
solutions to dynamic crack propagation for all velocity regimes. For a mode I crack,
both the energy release rate and the stress singularity vanish for all crack velocities in
excess of the Rayleigh wave speed. The implication is that there is no possibility for a
mode I crack to propagate at a velocity greater than the Rayleigh wave speed. A mode
II crack behaves in a similar manner to a mode I crack in the subsonic velocity range;
i.e. the energy release rate monotonically decreases to zero at the Rayleigh wave speed
1
We dedicate this work to Jim Rice and John Hutchinson who have provided more than three decades
of intellectual leadership on fracture research.
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
2115
and remains zero between the Rayleigh and shear wave speeds. Also, in both cases, the
order of the stress singularity is equal to 1}2 below the Rayleigh wave speed and zero
between the Rayleigh and shear wave speeds. However, the predictions for the two
loading modes dier for crack velocities greater than the shear wave speed. In contrast
to mode I where the order of the stress singularity remains zero for speeds greater than
the shear wave speed, the order for mode II becomes positive, indicating that only shear
cracks exhibit strong stress concentration in the intersonic velocity range. The energy
release rate is zero for a mode I crack but is positive for a mode II crack over the
entire range of intersonic velocities, with a maximum value near a special speed

2c
s
at which the crack stops radiating (Freund, 1979; Broberg, 1989; Gao et al., 1999).
Burridge et al. (1979), Freund (1979) and Simonov (1983) investigated various aspects
of intersonic crack growth such as the stability regime as well as stress singularity at
crack tip and at the radiating wavefronts. Broberg (1989, 1999) discussed that the
positive energy release rate for an intersonic mode II crack needs to be understood
from a cohesive view of fracture because in the conventional point singularity model,
the energy release rate is strictly zero beyond Rayleigh wave speed except for a mode
II crack propagating exactly at the radiation-free speed

2c
s
. Yu and Suo (2000) used
the cohesive model to study permissible velocity zones for intersonic crack growth
along a bimaterial interface.
Consider the excluded regions of crack velocity implied by the theoretical solutions.
For a mode I crack, any velocity larger than the Rayleigh wave speed is forbid-
den. Clearly, the limiting speed of a mode I crack is the Rayleigh wave speed. A
mode II crack faces a forbidden velocity zone between the Rayleigh and shear wave
speeds. Other than this forbidden zone, a mode II crack is allowed to propagate in
both sub-Rayleigh and super-shear velocity regimes. The limiting speed for a mode II
crack is the Rayleigh wave speed only if the forbidden velocity zone is impenetra-
ble. Is there a mechanism by which a shear crack could jump from sub-Rayleigh to
super-shear velocities? The rst hint at the answer to this question came from Bur-
ridges (1973) analysis of cracks growing self-similarly at the Rayleigh speed. He
found that, under mode II conditions, a positive peak of shear stress develops at the
shear wavefront ahead of the tip. In comparison, no such peak exists under mode I
conditions. These results suggest that a mode II crack at the Rayleigh wave speed
could induce secondary fracture ahead of the main crack tip. Is this a feasible mech-
anism for mode II cracks to jump over the forbidden velocity zone? Andrews (1976)
used a slip weakening model to investigate shear crack growth along a weak interface
and found that the shear crack approaching the Rayleigh wave speed indeed induces a
microcrack, or a daughter crack, that moves at speeds exceeding the shear wave speed.
Similar mechanisms were also reported in the cohesive surface nite element simula-
tions (Needleman, 2000), MD simulations (Abraham and Gao, 2000; Abraham, 2001)
and elastodynamic boundary integral simulations (Geubelle and Kubair, 2001) of in-
tersonic shear fracture. Geubelle and Kubair (2000) have also found intersonic speeds
without a daughter crack forming for a crack propagating in a pre-stressed specimen
under large mixed mode loading.
Evidence of shear crack propagation in excess of the shear wave speed has also been
provided from observations of shallow crustal earthquakes (Archuleta, 1982; Beroza and
2116
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
Spudich, 1988; Wald and Heaton, 1994). Among recent earthquakes where super-shear
rupture velocity has been reported is the M7.4 earthquake at Kocaeli, Turkey on August
17, 1999 (Ellsworth and Celebi, 1999). Although the experiments of Rosakis et al.
(1999) have shown convincingly that a shear dominated crack can propagate with
velocity up to the longitudinal wave speed, the question of whether such a crack has
been accelerated from a subsonic crack or is nucleated directly as an intersonic crack
has not been fully resolved by experiments.
Continuum models have provided signicant insights into the mechanism of inter-
sonic fracture. However, the origin of materials failure begins at the atomic scale where
atomic bonds are broken and atomic planes slide over each other. This is the ab initio
level of study for fracture. It would be interesting to test the validity and accuracy
of continuum models by direct atomic simulations of fracture. This issue is especially
important for intersonic cracks which are predicted to be of purely mode II character
based on continuum solutions. A mode II crack, according to linear elastic solutions,
has neither crack opening nor traction along the crack faces. Such a conguration could
not be comprehended from an atomistic point of view because atomic planes subject to
pure shear slide past one another via propagation of dislocations. There is thus some
conceptual diculty in distinguishing a mode II crack from an edge dislocation. Is
the intersonic crack actually an intersonic edge dislocation which was observed in MD
simulations (Hoover et al., 1977; Abraham et al., 1994; Gumbsch and Gao, 1999)?
This question has to be resolved by an atomistic study of fracture.
We have conducted a two-dimensional MD simulation to investigate the mechanisms
of shear crack propagation along a weak interface joining two harmonic crystals (Abra-
ham and Gao, 2000). In this paper, we apply the dynamic elasticity methods developed
by Freund (1990) to solve the crack propagation problem subject to identical geom-
etry and loading conditions in the MD simulation. We determine material properties
including Youngs modulus, Poisson ratio, wave speeds, surface energy and cohesive
strength from the interatomic potentials used in the atomic simulation. This precise
knowledge of material properties allows us to use the simulation results to test the
validity of continuum theories of fracture.
2. Atomic simulation results
The atomic simulation of intersonic shear fracture is based on molecular dynamics
which is a computational method (Abraham, 1986; Hoover, 1986; Allen and Tildesley,
1987) for predicting the motion of a given number of atoms by numerically integrating
Newtons law, F = ma, for each atom. In the MD simulation, the mutual interactions
among atoms are described by a continuous potential function. The details of MD
simulations of intersonic shear fracture can be found in Abraham and Gao (2000)
and Abraham (2001). Here we briey discuss the most relevant results which will be
compared to the continuum analysis shortly.
In accordance with the objective of studying crack propagation along a weak inter-
face in a linear elastic isotropic solid, we consider a two-dimensional atomic lattice
characterized by a pair potential. Atoms across a weak interface line are assumed to
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
2117
interact according to the LJ potential
[(r) = 4
_
1
r
12

1
r
6
_
. (1)
Here and throughout the paper all results are expressed in terms of reduced units:
lengths are scaled by the value of the interatomic separation for which the LJ potential
is zero, energies are scaled by the depth of the minimum of the LJ potential, and
mass is scaled by the atomic mass. A cut-o distance equal to 2.5 is assumed for the
LJ potential. Atoms in the adjacent crystals are assumed to interact according to the
harmonic potential
[(r) =
1
2
k(r d)
2
(2)
between nearest neighbors. This corresponds to the classical ball-spring model of
solids. In two dimensions, the above potential forms a stable triangular lattice which
is isotropic under small deformation. To make the material elastically homogeneous
across the interface, the lattice and spring constants of the harmonic crystal are taken
to be the same as for the LJ potential
d = 2
1}6
, k = [

(d) = 72}2
1}3
. (3)
We note that the harmonic crystal has innite fracture strength due to linear interactions
among nearest neighbors. The only fracture path in this linear elastic homogeneous and
isotropic solid is along the interface which has a nite cohesive strength associated with
the LJ potential.
Fig. 1 shows the MD simulation of intersonic shear crack propagation. The total
dimension of the simulation system under study is a 2D slab of atoms with 1424
atoms along the horizontal length dening the x direction and with 712 atoms along
the vertical length dening the , direction. A horizontal slit of 200 atom distance is
cut midway along the left-hand vertical slab boundary. The 2D crystal has a triangular
lattice with the slit parallel to the close packed direction along which atomic spacing
is equal to the lattice constant 2
1}6
. The initial temperature is set to be zero, and
the simulation is conducted at constant energy. To study a shear dominated crack, a
shear strain rate of 2.5 10
4
and a tensile strain rate of 5 10
5
are imposed on the
outermost rows of atoms dening the opposing horizontal faces of the two-dimensional
slab. The top of the slab is moving up and to the left (or in the positive , and
negative x directions) and the bottom of the slab is moving down and to the right
(or in the negative , and positive x directions). The crack is of mixed mode but
dominantly shear. Linear velocity gradients are established across the slab initially.
Then the loading process proceeds with constant shear and tensile velocities along the
boundary, which leads to eventual failure of the material at the slit tip. The applied
strain rates remain constant during the simulation, and the simulation is continued until
the growing crack has traversed the total length of the slab. Simulations of a mode
I crack are conducted with the same geometrical setup except that only an opening
strain rate equal to 5 10
5
is imposed.
The histories of crack velocity under mode I and mode II loading are presented in
Fig 2. The triangles show representative positions of the crack under tensile (mode I)
2118
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
Fig. 1. A snap shot of MD simulation of intersonic crack propagation. The total dimension of the simulation
system is 1424712 atoms. A horizontal slit of 200 atom distance is cut midway along the left-hand vertical
slab boundary parallel to the close packed lattice direction. The initial temperature is set to be zero, and the
simulation is conducted at constant energy.
loading as a function of time. The mode I crack quickly approaches a constant velocity,
as indicated by the tip positions falling on a straight line with a slope of 4.83 equal to
the Rayleigh wave speed of the harmonic crystal. (The values of the wave speeds are
further discussed in the next section.) The mode I case shows that the crack velocity is
limited by the Rayleigh wave speed, consistent with the classical theories of fracture.
In comparison, the dots show representative recorded positions of the crack under shear
dominated (mode II) loading as a function of time. The mode II crack is initiated at
a critical time estimated to be 65 and quickly approaches a constant velocity. After
propagating at this constant velocity for a short while, the crack tip jumps to a higher
constant velocity, as indicated by the positions falling on two straight lines. The slope
of the rst straight line is calculated to be 4.83, or the Rayleigh wave speed of the
harmonic solid. The slope of the second straight line is calculated to be 8.97 which is
essentially the longitudinal sound speed. The time for this velocity jump is estimated
to be around 140.
The mechanism for the mode II crack jumping over the forbidden velocity zone
is shown in Fig. 3 where a series of color maps of the shear stress component o
x,
is
used to reveal the details of this process. This transition occurs by the nucleation of an
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
2119
Fig. 2. The history of crack velocity under mode I and mode II loading. Note that the mode I crack is quickly
accelerated to the Rayleigh wave speed and is limited by this speed. The mode II crack rst accelerates to
the Rayleigh wave speed, propagates at the Rayleigh speed for a while and then jumps to the longitudinal
sound speed.
intersonic daughter crack ahead of the mother crack travelling at the Rayleigh wave
speed. As the mother crack approaches the critical state of nucleation, the crack tip
region is asymmetrically distorted with a bulge on the right side of the crack face,
as shown in Fig. 3(a). The linear elastic solutions of dynamic crack tip eld (Fre-
und, 1990) predict that the opening displacements along the crack face are symmetric
with respect to the crack line under mode I loading and zero under mode II load-
ing. (There is only slip-like motion of crack surfaces under mode II.) According to
these solutions, the crack opening displacements should remain symmetric even un-
der mixed mode conditions, which is inconsistent with the asymmetric distortion ob-
served in our simulation. It appears that the scale of asymmetry is too large to be
explained by the asymmetry of lattice. While the origin of such discrepancy between
theory and simulation is not entirely clear, we believe that the crack tip distortion
is a consequence of geometrical nonlinearity which causes local rotation of the crack
tip with respect to the interface. The asymmetric shear stress distribution near the
crack tip is also at odds with the mode II crack analysis of Burridge (1973) and
Andrews (1976), although the nucleation of the daughter crack is very reminiscent
of what they have proposed based on continuum analysis. These authors discovered
that there is a positive peak of shear stress at a distance ahead of a shear crack
moving at the Rayleigh speed and this stress peak may precipitate an intersonic micro-
crack. Fig. 3(b)(f) show the detailed process of the birth of the intersonic daughter
crack. It is seen that a sharp intersonic crack is nucleated at a small distance esti-
mated to be around 22 ahead of the mother crack. Color maps of atomic velocity
(not shown here) show transverse Mach cones near the daughter crack. The angle of
the Mach cone shows that the velocity of the daughter crack is consistent with the
longitudinal wave speed. As the daughter crack moves ahead, the mother crack can
still be seen as a surface bulge (Fig. 1) which trails behind at the Rayleigh wave
speed.
2120
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
Fig. 3. The nucleation of intersonic daughter crack at the mother crack. The gures represent a progression
in time from left to right and from top to bottom which are referenced as (a)(f). (a) The approach of the
critical state for the mother crack. Note the asymmetrically distorted crack tip region; (b) the conguration
of mother crack just before the intersonic crack is born; (c) the birth of the intersonic crack. A very sharp
slit is born ahead of the tip of the mother crack; (df) The daughter crack joins the mother crack and quickly
approaches the longitudinal sound speed. The colorbar shows the color map for the shear stress component
o
x,
. The bulged mother crack is still propagating at the Rayleigh wave speed. The dimensions of the solid
region shown are approximately 90 120 (in reduced units).
The MD simulations demonstrate intersonic crack propagation and the existence of
a motherdaughter crack mechanism for a subsonic shear crack to jump over the
forbidden velocity zone. This mechanism is reminiscent of the BurridgeAndrew mech-
anism based on continuum theories, although the continuum description cannot provide
an ab initio description for crack formation and the details of crack tip distortion are
not consistent with the continuum solutions. The birth of the daughter crack cannot be
characterized by a critical energy release rate or a critical stress intensity factor near
the mother crack because both these quantities vanish at the Rayleigh wave speed. It
seems that the only possible mechanism by which the daughter crack can be nucleated
is by the nite stress peak ahead of the mother crack, and along the weak bonding
line, as measured in the stress eld and discussed by Burridge (1973).
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
2121
3. Material properties
Material properties of importance to continuum descriptions of fracture include
Youngs modulus, Poisson ratio, elastic wave speeds, surface energy and cohesive
strength. In comparison with a laboratory fracture experiment, atomistic simulations
have the advantage of providing a precise knowledge of these material properties from
the interatomic potential. They are listed below in reduced units.
3.1. Elastic constants
A two-dimensional triangular lattice behaves as a plane stress elastic sheet with the
following moduli constants (Gao, 1996):
j =

3
4
k =
18

3
2
1}3
, E =
2

3
k =
144
2
1}3

3
, v = 1}3, (4)
where j is the shear modulus, E the Youngs modulus and v the Poisson ratio.
3.2. Elastic wave speeds
With the atomic mass taken as the unit of mass, the triangular lattice has the density
j =
2
2
1}3

3
. (5)
It follows that the longitudinal, shear and Rayleigh wave speeds are
c
d
=

3j
j
= 9, c
s
=
_
j
j
= 5.20, c
R
0.93c
s
= 4.83, (6)
respectively.
3.3. Fracture surface energy
The fracture surface energy of the material is dened as the energy consumed in
breaking atomic bonds as crack grows. For the MD simulations described in Section
2, the crack is parallel to the close-packed direction and atoms across the interface
interact according to the LJ potential with a cut-o distance equal to 2.5. Accounting
for all the atomic interactions, four atomic bonds (2 between nearest neighbors and
2 between next nearest neighbors) are snapped per atom in the fracture process. The
fracture surface energy is dened as half of the energy stored in these bonds and is
equal to
=
[(d) + [(

3d)
d
= 0.956. (7)
3.4. Cohesive strength
We are interested in the cohesive strength of the weak interface under shear domi-
nated loading. This strength is calculated as follows.
2122
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
Fig. 4. The conguration of atomic bonds across the interface. Cohesive failure is dened as the state when
one of the nearest neighbor bonds reaches the maximum bond force under applied shear stress t
int
and
tensile stress o
int
.
The cohesive failure of a single atomic bond is dened as the state when the inter-
active force between two atoms reaches the maximum. This corresponds to
[

(d
m
) = 0, (8)
where
d
m
= (26}7)
1}6
(9)
is the critical bond length at failure. We denote the cohesive strength of a single bond
as
[
m
= [

(d
m
) = 24
_

2
d
13
m
+
1
d
7
m
_
= 2.396. (10)
Fig. 4 shows the atomic bond conguration when a primary bond reaches cohesive
failure. Balance of forces parallel and normal to the interface leads to the following
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
2123
equations:
t
int
d = [
m
cos 0 + [
:
cos : [
1
cos :
1
+ [
2
cos :
2
,
o
int
d = [
m
sin 0 [
:
sin : + [
1
sin :
1
+ [
2
sin :
2
, (11)
where t
int
and o
int
are the shear and normal stresses along the interface at cohesive
limit; forces and angles of the dierent atomic bonds are marked in Fig. 4. Eliminating
[
:
from Eqs. (11) and using trignometric identities
d
sin(0 + :)
=
d
m
sin :
,
2d
sin(: + :
2
)
=
d
2
sin :
,
d
sin(: :
1
)
=
d
1
sin :
(12)
lead to
t
int
+ o
int
cos :
sin :
=
[
m
d
m
+ 2
[
2
d
1

[
1
d
1
. (13)
To a good approximation, we can take the corresponding values in the equilibrium
conguration
: = 60

, d
1
= d
2
=

3d, [
1
= [
2
= [

3d) (14)
for the distorted triangular lattice. The cohesive failure criterion of the interface is thus
t
int
+
o
int

3
=
[
m
d
m
+
[

3d)

3d
= 2.039. (15)
Note that there is a strong coupling between shear and tensile stresses during cohesive
failure.
4. Continuum analysis of crack initiation and propagation
The atomistic simulation discussed in Section 2 has clearly demonstrated that a
shear dominated crack could initiate and propagate along a weak plane in an otherwise
homogeneous material. Under the dynamic loadings, the crack tip remained stationary
for a while before it started to accelerate and propagated at the Rayleigh wave speed
c
R
. As the applied load continued to increase, a daughter crack (microcrack) was
nucleated at a nite distance ahead of the crack tip, and propagated intersonically
(faster than the shear wave speed) within the crack plane. This entire process of crack
initiation and propagation is studied in this section via a transient, continuum analysis
of dynamic fracture. It is shown that the continuum analysis, in conjunction with the
Grith criterion, can determine the critical time for crack initiation rather accurately.
The location and time at which the daughter crack is nucleated are also determined by
the continuum analysis, together with a cohesive strength criterion. Predictions based
on continuum analysis agree very well with those from atomistic simulations.
An innite plane-stress solid containing a semi-innite crack on the negative x
1
-axis
is subjected to constant remote shear stress rate t
0
and tensile stress rate o
0
normal
to the crack. Consistent with the atomistic simulations, an initial velocity eld at time
t = 0 corresponding to the constant remote stress rates is imposed such that there are
no waves coming from the remote eld. The solid is linear elastic and isotropic, with
the shear modulus j, Poisson ratio v, and Youngs modulus E = 2j(1 + v).
2124
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
The deformation eld can be decomposed into the following two sub-problems. First,
a uniform deformation eld corresponding to constant shear stress rate t
0
and normal
stress rate o
0
in the same solid but without the crack; the initial velocity eld is
consistent with t
0
and o
0
such that there are no waves from the remote eld. The
non-zero stresses are
o
12
= t
0
t, o
22
= o
0
t. (16)
The second sub-problem has constant shear and normal traction rates, t
0
and o
0
, im-
posed on the entire crack faces (including the new ones generated by crack propagation)
in order to negate the crack-face tractions from the rst sub-problem. There is no initial
velocity eld. The crack tip remains stationary until a critical time, t =t
init
, is reached
at which the Grith criterion is met. The crack tip then propagates in the crack plane
at the Rayleigh wave speed c
R
, consistent with the atomistic simulation discussed in
Section 2. It should be pointed out that stresses are not singular (!) near a crack tip
propagating at the Rayleigh wave speed c
R
. Moreover, the shear stress has a peak that
occurs at a nite distance ahead of the crack tip. Once the peak stress reaches the
cohesive strength of the solid, the daughter crack is nucleated.
4.1. Grith criterion and crack initiation
We study rst the critical time for crack initiation, t
init
, at which the macroscopic
crack tip starts to propagate. The stress intensity factors for a stationary plane-stress
crack tip subjected to constant remote shear and normal traction rates t
0
and o
0
on the
crack faces are (e.g., Freund, 1990)
K
II
(t) =
4
3
t
0
_
2(1 + v)c
s
t
3

(17)
and
K
I
(t) =
4
3
o
0

2 2v(1 + v)c
s
t
3

. (18)
The corresponding plane-stress crack tip energy release rate is
G =
1
E
(K
2
II
+ K
2
I
) =
16c
s
9j
_
t
2
0
+
_
1 v
2
o
2
0
_
t
3
. (19)
The Grith criterion predicts that the crack tip starts to propagate when the crack
tip energy release rate reaches twice the surface energy, 2,
G = 2. (20)
The critical time for crack initiation, at which the macroscopic crack tip starts to
propagate, is determined from (19) and (20) as
t
init
=
_
_
9j
8c
s
( t
2
0
+
_
1v
2
o
2
0
)
_
_
1}3
. (21)
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
2125
It predicts that the crack initiation time is proportional to the surface energy via
1}3
,
and is inversely proportional to the applied shear and normal stress rates via t
2}3
0
and
o
2}3
0
in modes II and I, respectively.
The elastic constants, wave speeds and fracture surface energy are given, respec-
tively, in (4), (6) and (7) for the LJ potential. The (engineering) shear strain rate and
normal strain rate are
0
=0.00025 and c
0
=0.00005 in the atomistic simulations. The
corresponding shear stress rate t
0
and normal stress rate o
0
are determined from the
linear elastic constitutive relation. These give the critical time for crack initiation in
(21) as
t
init
= 70.3 (22)
in reduced unit, which is in good agreement with the corresponding result t
MD
init
=65 of
the MD simulations. This analysis indicates that the Grith criterion holds even down
to the atomic scale.
4.2. Crack propagation: mode II analysis
The continuum analysis becomes much more dicult after the crack tip propagates at
the Rayleigh wave speed c
R
after time t =t
init
. Let (x
1
, x
2
) be the Cartesian coordinates
centered at the initial (stationary) crack tip, and (
1
,
2
) be coordinates moving with
the crack tip, i.e.,
1
= x
1
c
R
t and
2
= x
2
. The in-plane stresses and displacements
are denoted by o
:[
and u
[
(:, [ =1, 2), respectively. There is no initial deformation at
time t = 0 (even though there is an initial velocity eld), i.e.,
u
[
(t = 0) = 0. (23)
We rst study the second sub-problem stated above with the crack-face shear loading
(i.e., mode II), while the mode I solution is given in the next sub-section.
Only the upper half plane (x
2
0) is analyzed due to symmetry. The boundary
conditions in the mode II problem are
o
22
(x
2
= 0) = 0, x
1
+, (24)
u
1
(x
2
= 0) = 0, max[0, c
R
(t t
init
)] x
1
+ (25)
and
o
12
(x
2
= 0) = t
0
t, x
1
max[0, c
R
(t t
init
)], (26)
where max[0, c
R
(tt
init
)] represents the crack tip location. We follow the method devel-
oped by Freund (1990) to solve this fully transient dynamic fracture problem involving
both the dynamic crack-face loadings and the crack propagation. Laplace transform is
applied with respect to time, followed by the two-sided Laplace transform with respect
to coordinate
1
. The WienerHopf technique is used to solve the transformed stress
tractions ahead of the crack tip and the transformed displacement on the crack face.
2126
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
The inverse Laplace transform, in conjunction with the Cagniardde Hoop method,
then gives the analytic solution following propagation,
We skip the details of the solution and present only the shear stress relevant to the
nucleation of daughter crack. The total shear stress ahead of the moving crack tip is the
sum of the uniform shear stress in (16) from the rst sub-problem and the nonuniform
shear stress t from the second sub-problem stated above,
o
12
(
1
0,
2
= 0) = t
0
t + t, t t
init
, (27)
where t is given by
t =
t
0
c
s
s
+
(0)
0
H
_
t t
init

1
c

c
R
__
(tt
init
)}
1
1}(c

c
R
)
_
(c
s
+ c
R
)r + 1
rs

(r)
F
1
(r)(t
1
r) dr +
t
0
c
s

0
H
_
t t
init

1
c

c
R
_

_
+
1}c
R
(w)Q(w,
1
, t)
_
(c
s
c
R
)w + 1
s
+
(w)
dw +
t
0
c
s
s
0+
(0)
c
2
R
2(
1
c
2
s

1
c
2

)
H
_
t

1
+ c
R
(t t
init
)

1
c

_ _
t}(
1
+c
R
(tt
init
))
max[
1
c

, (tt
init
)}(
1
+c
R
(tt
init
))]

c
s
r + 1
r(c
R
r + 1)s
0
(r)
F
2
(r)[t
1
r c
R
r(t t
init
)] dr. (28)
Here, H is the Heaviside step function; the constant
0
is given in terms of the Rayleigh
wave speed c
R
, shear wave speed c
s
and longitudinal wave speed c

by

0
=
2
c
2
R

1
c
2
R
c
2
s

1
c
2
R
c
2

_
1
c
2
s
c
2
R
+
1
c
2

c
2
R
_

2
c
2
R
c
2
s
_
2
c
2
R
c
2
s
_
. (29)
The functions s

and s
+
are given by
s

(r) = exp
_

_
1}(c
s
c
R
)
1}(c

c
R
)
tan
1
4c
4
s
p
2
[(p)|:(p)|
(2c
2
s
p
2
(c
R
p 1)
2
)
2
dp
p + r
_
, (30)
[(p) =
1
c
s
_
1 (c
s
c
R
)p
_
1 (c
s
+ c
R
)p, (31)
|:(p)| =
1
c

_
(c

c
R
)p 1
_
(c

+ c
R
)p 1. (32)
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
2127
Similarly, the function s
0
(r) in (28) is
s
0
(r) = exp
_
_

_
1}c
s
1}c

tan
1
4c
3
s
p
2
_
1 c
2
s
p
2
_
c
2

p
2
1
c

(2c
2
s
p
2
1)
2
dp
p + r
_
_
; (33)
the constant s
0+
(0) = s
0
(0), and therefore can be evaluated from (33); the functions
F
1
and F
2
in (28) are dened as
F
1
(r) =
1
(c
R
r + 1)
2
(2c
R
r + 1)
_
4r
2
|:(r)|
(2c
2
s
r
2
(c
R
r + 1)
2
)
2
c
4
s
|[(r)|
H
_
r
1
c
s
c
R
__
, (34)
F
2
(r) =
4
c

r
2
_
c
2

r
2
1
(2c
2
s
r
2
1)
2
c
3
s
_
c
2
s
r
2
1
H
_
r
1
c
s
_
; (35)
and nally, the functions and Q in (28) are given by
(w) =
1
c
s
s
0+
(0)
c
2
R
2(1}c
2
s
1}c
2

)
_
w}(wc
R
1)
1}c

c
s
r + 1
r(c
R
r + 1)s
0
(r)
F
2
(r) dr, (36)
Q(w,
1
, t)
=
_
(tt
init
)}
1
max(1}(c

c
R
),(tc
R
wt
init
)}(
1
+c
R
t
init
))
_
(c
s
+ c
R
)r + 1
(r + w)
2
s

(r)
F
1
d(r)

__
s

+
(w)
s
+
(w)

1
2
(c
s
c
R
)
(c
s
c
R
)w + 1
_
r + w
c
R
(r + w) 1
(t
1
r c
R
t
init
(r + w))
+
2c
R
(r + w) 1
(c
R
(r + w) 1)
2
(t t
init

1
r) t
init
_
dr. (37)
The shear stress distribution ahead of the moving crack tip in (27) is shown in
Fig. 5 at 1.5 times the initiation time for crack propagation, i.e., t =1.5t
init
, where the
shear stress is normalized by the constant shear stress traction rate t
0
and the initiation
time t
init
, and the distance
1
0 to the moving crack tip is normalized by the shear
wave speed c
s
and t
init
. It is clearly observed that the shear stress has a very sharp
peak ahead of the moving crack tip. This sharp peak occurs at the shear wavefront,
i.e., at a distance of (c
s
c
R
)(t t
init
) to the moving crack tip. In fact, it can be
shown from (27) that, for any time t t
init
, the peak stress always occurs at the shear
wavefront.
2128
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
Fig. 5. The distribution of shear stress ahead of the mother crack moving at the Rayleigh wave speed. Note
that there is a stress peak at the shear wavefront.
Based on this conclusion, the peak stress t
peak
ahead of the moving crack tip can
be obtained from (27) by substituting
1
= (c
s
c
R
)(t t
init
). This gives
t
peak
= t
0
t +
t
0
c
s
s
+
(0)
0
_
1}(c
s
c
R
)
1}(c

c
R
)
_
(c
s
+ c
R
)r + 1
rs

(r)
F
1
(r)[t (c
s
c
R
)r(t t
init
)] dr
+
t
0
c
s

0
_
+
1}c
R
(w)Q
peak
(w, t)
_
(c
s
c
R
)w + 1
s
+
(w)
dw
+
t
0
c
s
s
0+
(0)
c
2
R
2(1}c
2
s
1}c
2

_
t}c
s
(tt
init
)
1}c
s

c
s
r + 1
r(c
R
r + 1)s
0
(r)
F
2
(r) [t c
s
r(t t
init
)] dr, (38)
where the constants and functions are the same as in (29)(37), except that the func-
tion F
1
takes a much simpler form of F
1
(r) = 4r
2
|:(r)|}(c
R
r + 1)
2
(2c
R
r + 1), and
the function Q
peak
is Q in (37) evaluated at the shear wavefront, i.e., Q
peak
(w, t) =
Q(w, (c
s
c
R
)(t t
init
), t).
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
2129
4.3. Crack propagation: mode I analysis
The mode I solution can be obtained following the same approach. Only the normal
stress at the shear wavefront ahead of the moving crack tip, o = o
22
[
1
= (c
s
c
R
)
(t t
init
),
2
= 0], is presented in the following:
o = o
0
t +
o
0
c

s
+
(0)
0
_
1}(c
s
c
R
)
1}(c

c
R
)
_
(c

+ c
R
)r + 1
rs

(r)
F
(I )
1
(r)[t (c
s
c
R
)r(t t
init
)] dr +
o
0
c

_
+
1}c
R

(I )
(w)Q
(I )
(w, t)
_
(c

c
R
)w + 1
s
+
(w)
dw +
o
0
c

s
0+
(0)
c
2
R
2(1}c
2
s
1}c
2

_
t}c
s
(tt
init
)
1}c
s

r + 1
r(c
R
r + 1)s
0
(r)
F
(I )
2
(r)[t c
s
r(t t
init
)] dr. (39)
Here the constant
0
and functions s

, [, |:| and s
0
remain the same as in
(29)(33), while other functions are modied from (34)(37) for mode I as
F
(I )
1
(r) =
(2c
2
s
r
2
(c
R
r + 1)
2
)
2
c
4
s
(c
R
r + 1)
2
(2c
R
r + 1)|:(r)|
, (40)
F
(I )
2
(r) =
4
c
s
r
2
_
c
2
s
r
2
1H
_
r
1
c
s
_

(2c
2
s
r
2
1)
2
c
4
s
_
c
2

r
2
1
, (41)

(I )
(w) =
1
c

s
0+
(0)
c
2
R
2(1}c
2
s
1}c
2

)
_
w}(wc
R
1)
1}c

r + 1
r(c
R
r + 1)s
0
(r)
F
(I )
2
(r) dr
(42)
and
Q
(I )
(w, t) =
_
1
c
s
c
R
max(
1
c

c
R
,
tc
R
wt
init
(c
s
c
R
)(tt
init
)+c
R
t
init
)
_
(c

+ c
R
)r + 1
(r + w)
2
s

(r)
F
(I )
1
(r)

__
s

+
(w)
s
+
(w)
(r + w)
1
2
(c

c
R
)(r + w)
(c

c
R
)w + 1
+
2c
R
(r + w) 1
c
R
(r + w) 1
_

_
1 (c
s
c
R
)r
c
R
(r + w) 1
(t t
init
) t
init
_
+
c
R
(r + w)
c
R
(r + w) 1
t
init
_
dr.
(43)
2130
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
4.4. Cohesive strength criterion and the nucleation of daughter crack
We use the cohesive strength criterion in (15) to determine the nucleation of the
daughter crack. The maximum shear stress in (15) is the peak shear stress t
peak
at the
shear wavefront in (38), while the corresponding normal stress in (15) is replaced by
o also at the shear wavefront in (39). From the dimensional ground, the peak shear
stress in (38) and normal stress in (39) take the form
t
peak
= t
0
t
init
[
II
_
t
t
init
_
, o = o
0
t
init
[
I
_
t
t
init
_
, (44)
where [
II
and [
I
are nondimensional functions of the normalized time t}t
init
obtained
from (38) and (39), respectively. The substitution of (44) into the cohesive strength
criterion (15) yields the equation to determine the critical time t
nucl
for the nucleation
of the daughter crack
t
0
[
II
_
t
nucl
t
init
_
+
o
0

3
[
I
_
t
nucl
t
init
_
=
2.039
t
init
. (45)
For the material properties given in Section 3 and the applied shear and normal stress
rates t
0
and o
0
calculated from the (engineering) shear and normal strain rates
0
=
0.00025 and c
0
= 0.00005 in atomistic simulations via the linear elastic constitutive
relation, (45) predicts the ratio of critical time for daughter crack nucleation, t
nucl
, to
that for crack initiation, t
init
, as t
nucl
}t
init
=1.71. Based on t
init
=70.3 in (22), the critical
time for the nucleation of the daughter crack is predicted as
t
nucl
= 120 (46)
in the reduced unit, which is in reasonable agreement with the counterpart of t
MD
nucl
=140
from atomistic simulations.
The corresponding location
nucl
at which the daughter crack is nucleated ahead of
the moving crack tip is the shear wavefront and can be determined as

nucl
= (c
s
c
R
)(t
nucl
t
init
) = 18.2 (47)
in reduced unit, which is once again in reasonable agreement with the estimate of

MD
nucl
=22 from atomistic simulations. This indicates that the cohesive strength criterion
seems to govern the nucleation of the daughter crack even down to the atomic scale,
leading to intersonic crack propagation.
5. Conclusions
We have studied intersonic shear crack propagation along a weak interface by both
molecular dynamics and continuum elastodynamics. The problem selected is a block
of linearly isotropic, plane-stress elastic solid consisting of a two-dimensional triangu-
lar atomic lattice with pair interatomic potential loaded by constant shearing velocity
along the boundary. The fracture process revealed by MD simulations shows the fol-
lowing sequence of events. The initial crack starts to grow at a critical time after the
loading process begins. It quickly accelerates to the Rayleigh wave speed and, after
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
2131
Table 1
Comparison between continuum predictions of critical time and length scales for intersonic crack propagation.
t
mother
is the time at which the mother crack begins to grow; t
daughter
is the time when the daughter crack
is born; L
daughter
is the location ahead of the mother crack where the daughter crack is born
Continuum analysis Atomistic simulation
Time for crack initiation 70.3 65
Time for daughter crack nucleation 120 140
Location of daughter crack nucleation 18.2 22
propagating at this speed for a short time period, nucleates an intersonic daughter crack
which immediately jumps to the longitudinal wave speed. The daughter crack emerges
at a critical distance ahead of the mother crack. We solve the continuum elastody-
namic problem of the same crack geometry under the same loading history to test if
the continuum analysis can correctly predict the length and time scales observed in the
atomic simulations. We assume that the crack initiation is governed by the Grith cri-
terion while the nucleation of the daughter crack is governed by the BurridgeAndrew
mechanism of cohesive failure by a peak of shear stress ahead of the crack tip. We
determine material properties including elastic constants, elastic wave speeds, fracture
surface energy and cohesive strength from the interatomic potential used in the atomic
simulations.
Table 1 summarizes the quantitative comparison between the continuum mechanics
predictions and the corresponding results of the atomic simulations. It is seen that the
critical time for initial crack growth predicted by the Grith criterion agrees with
the simulation results within 10%. Also, we nd remarkably good agreement between
continuum analysis and atomic simulations for the time and location of the nucleation
of the daughter crack. From this comparison, we conclude that continuum mechanics
cannot only provide qualitatively useful insights into the mechanisms of intersonic
shear crack propagation, but also gives quantitatively correct predictions for the times
and locations of critical atomistic events. Eective linking between continuum and
atomistic methods is expected to be a powerful way of studying a wide variety of
nanoscale dynamic phenomena.
Acknowledgements
The work of HG was supported by the NSF through Grant CMS-9820988. The work
of YH was supported by the NSF through Grant CMS-9983779 and NSF of China. FFA
acknowledges support from the USAF Wright-Patterson High Performance Computing
Center and the San Diego Supercomputer Center which has its major funding from
NSF. FFA wishes to thank DODs HPCMO for a Grand Challenge award.
References
Abraham, F.F., 1986. Computational statistical mechanics: methodology, applications and supercomputing.
Adv. Phys. 35, 1111.
2132
H. Gao et al. / J. Mech. Phys. Solids 49 (2001) 21132132
Abraham, F.F., Brodbeck, D., Rafey, R.A., Rudge, W.E., 1994. Instability dynamics of fracture: a computer
simulation investigation. Phys. Rev. Lett. 73, 272275.
Abraham, F.F., Brodbeck, D., Rudge, W.E., Broughton, J.Q., Schneider, D., Land, B., Lifka, D., Gerner, J.,
Rosenkrantz, M., Skovira, J., Gao, H., 1998. Ab initio dynamics of rapid fracture. Modeling Simulation
Mater. Sci. Eng. 6, 639670.
Abraham, F.F., 2001. Fragile atland. J. Mech. Physics Solids, in the press.
Abraham, F.F., Gao, H., 2000. How fast can cracks propagate? Phys. Rev. Lett. 84, 31133116.
Allen, M.P., Tildesley, D.J., 1987. Computer Simulation of Liquids. Clarendon Press, Oxford, England.
Andrews, D.J., 1976. Rupture velocity of plane strain shear cracks. J. Geophys. Res. 81, 56795687.
Archuleta, R.J., 1982. Analysis of near source static and dynamic measurements from the 1979 Imperial
Valley earthquake. Bull. Seismol. Soc. Am. 72, 19271956.
Beroza, G.C., Spudich, P., 1988. Linearized inversion for fault rupture behaviorapplication to the 1984
Morgan-Hill, California, Earthquake. J. Geophys. Res. 93, 62756296.
Broberg, K.B., 1989. The near-tip eld at high crack velocities. Int. J. Fracture 39, 113.
Broberg, K.B., 1999. Cracks and Fracture. Academic Press, San Diego.
Burridge, R., 1973. Admissible speeds for plane-strain self-similar shear cracks with friction but lacking
cohesion. Geophys. J. Roy. Astron. Soc. 35, 439455.
Burridge, R., Conn, G., Freund, L.B., 1979. The stability of a plane strain shear crack with nite cohesive
force running at intersonic speeds. J. Geophys. Res. 84, 22102222.
Ellsworth, W.L., Celebi, M., 1999. Near eld displacement time histories of the M 7.4 Kocaeli (Izimit),
Turky, Earthquake of August 17, 1999. EOS Trans. Am. Geophys. Union (AGU 1999 Fall Meeting). 80
(46), F648.
Freund, L.B., 1979. The mechanics of dynamic shear crack propagation. J. Geophysics Res. 84, 21992209.
Freund, L.B., 1990. Dynamic Fracture Mechanics. Cambridge University Press, Cambridge, England.
Gao, H., 1996. A theory of local limiting speed in dynamic fracture. J. Mech. Phys. Solids 44, 14531474.
Gao, H., Huang, Y., Gumbsch, P., Rosakis, A.J., 1999. On radiation-free transonic motion of cracks and
dislocations. J. Mech. Phys. Solids 47, 19411961.
Geubelle, P.H., Kubair, D., 2001. Intersonic crack propagation in homogeneous media: Numerical analysis.
J. Mech. Phys. Solids 49, 571587.
Gumbsch, P., Gao, H., 1999. Dislocations faster than the speed of sound. Science 283, 965968.
Hoover, W.G., Hoover, N.E., Moss, W.C., 1977. Steady-state dislocation-motion via molecular-dynamics.
Phys. Lett. A 63, 324326.
Hoover, W.G., 1986. Molecular Dynamics. Springer, Berlin.
Needleman, A., 2000. An analysis of intersonic crack growth under shear loading. J. Appl. Mech. 66,
847857.
Rosakis, A.J., Samudrala, O., Coker, D., 1999. Cracks faster than the shear wave speed. Science 284,
13371340.
Simonov, I.V., 1983. Behavior of solutions of dynamic problems in the neighborhood of the edge of a cut
moving at transonic speed in an elastic medium. Mech. Solids (Mech. Tverdogo Tela) 18, 100106.
Wald, D.J, Heaton, T.H., 1994. Spatial and temporal distribution of slip for the 1992 Landers, California,
Earthquake. Bull. Seismol. Soc. Am. 84, 668691.
Yu, H.H., Suo, Z., 2000. Intersonic crack growth on an interface. Proc. Roy. Soc. Lond. A456, 223246.