Professional Documents
Culture Documents
Michael P. Marder
Center for Nonlinear Dynamics and Department of Physics
The University of Texas at Austin
Austin Texas, 78712, USA
February 7, 2003
Contents
1 Continuum Theories of Fracture: Scaling 2
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Scaling theory I: Things fall apart . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Scaling Theory II: All together . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Scaling theory III: Cracks at last . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 The scaling of dynamic fracture . . . . . . . . . . . . . . . . . . . . . . . 6
3 Incorporating Plasticity 25
1
4.6 Equations solved in terms of a single mass point on crack line: . . . . . . . 35
4.7 Relation between and . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.8 Phonon emission: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.9 Forbidden velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.10 Nonlinear Instabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.11 The connection to the Yoffe instability . . . . . . . . . . . . . . . . . . . . 45
4.12 Generalizing to Mode I: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2
The lectures will begin by establishing the basic stability criteria that determine when
fractures to begin moving. They will proceed to simple scaling theories for rapid crack
motion, and then to a brief description of linear elastic fracture mechanics, an ingenious
framework that allows a huge number of practical questions about fractures to be answered
without needing to resolve how they behave near the tip. However, linear elastic fracture
mechanics is not complete, and the lectures will continue with a discussion of current
attempts to provide the missing information, including continuum theories of crack tips,
atomic theories of crack tips, and techniques to consider cracks simultaneously at many
scales.
4G
h 14cm (1)
g
So every block of stone more than 20 cm tall is unstable under its own weight. A similar
estimate applies to steel, concrete, or the bones in our legs. Almost every feature in the
solid world around us is out of mechanical equilibrium.
Mountains, buildings, and even our own bodies should fall apart in the long run under
the influence of gravity. Yet they are apparently very stable, something which can only be
explained if strong energy barriers keep them intact. The next task is to estimate the size of
these barriers.
F EA
L
(2)
3
F
dL
L
L
A
F
Figure 1: Mechanically stable configurations are often far from their lowest energy state.
For example, a solid completely free of flaws would only pull apart when all bonds as long
a plane snapped simultaneously, despite the fact that it would generally be energetically
advantageous for the solid to be split.
Material Young’s Modulus Theoretical Strength Practical Strength Practical/Theoretical
GPa GPa GPa
Iron 195-205 43-56 0.3 0.006
Copper 110-130 24-55 0.2 0.005
Titanium 110 31 0.3 0.009
Silicon 110-160 45 0.7 0.01
Glass 70 37 0.4 0.01
Plexiglas 3.6 3 0.05 0.01
Table 1: The experimental strength of a number of materials in polycrystalline or amor-
phous form, compared to their theoretical strength, from [1] and [2]
where E is the material’s Young’s modulus, so the force per area needed to reach the break-
ing point is around
FA E5
(3)
As shown in Table 1, this estimate is in error by orders of magnitude. The problem does
not lie in the crude estimates used to obtain the forces at which bonds separate, but in
the conception of the calculation. The first scaling argument greatly underestimated the
practical resistance of solid bodies to separation, while this one greatly overestimates it.
The only way to uncover correct orders of magnitude is to account for the actual dynamical
mode by which brittle solids fail, which is by cracking.
4
taneous failure of a macroscopic plane’s worth of bonds leads to a huge overestimation of
practical strength. To obtain even the correct order of magnitude in describing the resis-
tance of solids to failure, one must consider the presence of cracks.
A crack is a region of consecutive broken bonds. A crack in an otherwise perfect
material creates a stress singularity at its tip. For an atomically sharp crack tip, a single
crack a few microns long is sufficient to explain the large gaps between the theoretical
and experimental material strengths as shown in Table 1. The theoretical strength of glass
fibers, for example, can be approached by means of acid etching of the fiber surface. The
etching process serves to remove any initial microscopic flaws along the glass surface. By
removing its initial flaws, a 1mm2 glass fiber can lift a piano (but during the lifting process,
I wouldn’t advise standing under the piano). The stress singularity that develops at the tip of
a crack serves to focus the energy that is stored in the surrounding material and efficiently
use it to break one bond after another. Thus, the continuous advance of the crack tip, or
crack propagation, provides an efficient means to overcome the energy barrier between two
equilibrium states of the system having different amounts of mechanical energy.
L
dl
L y
dl
x
Figure 2: The upper surface of a strip of height L and thickness is rigidly displaced
upwards by distance . A crack is cut through the center of the strip, and relieves all
stresses in its wake. When the crack moves distance dl from (A) to (B), the net effect is to
transfer length dl of strained material into a length dl of unstrained material.
Consider the crack moving in a thin strip depicted in Figure 2. Far to the right of the
crack tip the material is under tension, and stores an elastic energy right per area
right
1 2E
(4)
2 L
Far to the left of the crack, the material is completely relaxed, and therefore if the bound-
aries of the system are rigid as the crack proceeds in a steady fashion, the energy dissipated
by crack motion must be
1 2E
G (5)
2 L
per unit area crack advance. The reason for this assertion is that when the crack moves
ahead in a steady fashion the stress and strain fields around it do not change; as shown
in Figure 2, the fields translate in the direction of crack motion, and the change in energy
5
comes from the transformation of stressed material ahead of the crack to unstressed material
behind.
This conclusion rests upon symmetry, and does not even demand that strains ahead
of the crack tip to be so small that linear mechanics be applicable. Steady velocity
and a corresponding energy flux G are achieved in the long time limit, the natural scale
in experiments as well as in analytical calculations, where the crack tip reaches dynamic
equilibrium with waves reflecting from top and bottom boundaries. That is, the return
of acoustic waves from the system boundaries actually simplifies the task of understanding
energy balance in the thin strip geometry. According to fracture mechanics, the relationship
between energy flowing to a crack tip and crack velocity is, for a given lattice direction,
universal. Having obtained the relation in a strip, one knows it for any of the vast range
of geometries to which fracture mechanics is applicable, such as a long crack in a large
plate [3].
l
The first analysis of rapid fracture was carried out by Mott [4], and then slightly im-
proved by Dulaney and Brace [5]. It is a dimensional analysis which clarifies the basic
physical processes, despite being wrong in many details, and consists of writing down an
energy balance equation for crack motion.
Consider a crack of length l t growing at rate t in a very large plate where a stress
is applied at the far boundaries of the system, as shown in Figure 3. When the crack
extends, its faces separate, causing the plate to relax within a circular region centered on
6
the middle of the crack and with diameter of order l. The kinetic energy involved in moving
a region of this size is M 2 2, where M is the total mass, and is a characteristic velocity.
Since the mass of material that moves is proportional to l 2 , the kinetic energy is guessed to
be of the form
KE CK l 2 2 (6)
The region of material that moves is also the region from which elastic potential energy is
being released as the crack opens. Therefore, the potential energy gained in releasing stress
is guessed to be of the form
PE "! CP l 2 (7)
These guesses are correct for slowly moving cracks, but fail qualitatively as the crack ve-
locity approaches the speed of sound, in which case both kinetic and potential energies
diverge. This divergence will be demonstrated later, but for the moment, let us proceed
fearlessly. The final process contributing to the energy balance equation is the creation of
new crack surfaces, which takes energy # l. The fracture energy # accounts not only for
the minimal energy needed to snap bonds, but also for any other dissipative processes that
may be needed in order for the fracture to progress; it is often orders of magnitude greater
than the thermodynamic surface energy. However, for the moment, the only important fact
is that creating new surface scales as the length l of the crack. So the total energy of the
system containing a crack is given by
$ CK l 2 2 % qs l (8)
with
qs l &"! CP l 2 % # l (9)
Consider first the problem of quasi-static crack propagation. If a crack moves forward only
slowly, its kinetic energy will be negligible, so only the quasi-static part of the energy,
qs, will be important. For cracks where l is sufficiently small, the linear cost of fracture
energy is always greater than the quadratic gain of potential energy, and in fact such cracks
would heal and travel backwards if it were not for irreversible processes, such as oxidation
of the crack surface, which typically prevent it from happening. That the crack grows
at all is due to additional irreversible processes, sometimes chemical attack on the crack
tip, sometimes vibration or other irregular mechanical stress. It should be emphasized
that the system energy increases as a result of these processes. Eventually, at length l 0 ,
the energy gained by relieving elastic stresses in the body exceeds the cost of creating new
surface, and the crack becomes able to extend spontaneously. One sees that at l 0 , the energy
functional qs l has a quadratic maximum. The Griffith criterion for the onset of fracture
is that fracture occurs when the potential energy released per unit crack extension equals
the fracture energy, # . Thus fracture in this system will occur at the crack length l 0 where
d qs
dl
0
(10)
7
Quasistatic energy Eqs l0
Crack length l
Figure 4: The energy of a plate with a crack as a function of length. In the first part of its
history, the crack grows quasi-statically, and its energy increases. At l 0 the crack begins to
move rapidly, and energy is conserved.
8
2 Continuum fracture mechanics
In this section I will attempt to review briefly the background, basic formalism and under-
lying assumptions that form the body of continuum fracture mechanics. There are many
textbooks on the subject from many different points of view [8–14] , and I will only cover
a small subset of topics in them. I will first discuss the reduction of linear elasticity to two
dimensions. Some basic concepts common to both static and dynamic cracks will then be
discussed. These will include the creation of a singular stress field at the tip of a static
crack together with criteria for the onset and growth of a moving crack. I then turn to a
description of moving cracks. I will first describe the formalism used to describe a crack
moving at a given velocity and, using this formalism, look at the dependence of the near tip
stress fields as a function of the crack’s velocity. I then discuss criteria for determination
of a crack’s path and conclude the section by discussing a number of points that the theory
can not address.
9
?@<BADCE5;/F72GIH&7JAK<BLEMBHNGI/2APORQ
SUTWVYX[Z
.0/2143658727:9;1=<>5
Figure 5: Structure of fracture mechanics. The crack tip is surrounded by a region in which
the physics is unknown. Outside this process zone is a region in which elastic solutions
adopt a universal form.
The process zone: In the region immediately surrounding the crack tip, called the process
zone (or cohesive zone), all of the nonlinear dissipative processes that ultimately
allow a crack to move forward, are assumed to occur. Fracture mechanics avoids
any sort of detailed description of this zone, and simply posits that it will consume
some energy # per unit area of crack extension. The size of the process zone is
material dependent, ranging from nanometers in glass to microns in brittle polymers,
to centimeters or larger in ice near its melting point. The typical size of the process
zone can be estimated by using the radius at which an assumed linear elastic stress
field surrounding the crack tip would equal the yield stress of the material.
The universal elastic region: Everywhere outside of the process zone the response of the
material can be described by continuum linear elasticity. In the vicinity of the tip,
but outside of the process zone, the stress and strain fields adopt universal singular
forms which solely depend on the symmetry of the externally applied loads. In two
dimensions the singular fields surrounding the process zone are entirely described by
three constants, called stress intensity factors. The stress intensity factors incorporate
all of the information regarding the loading of the material and are related, as we shall
see, to the energy flux into the process zone. The larger the overall size of the body
in which the crack lives, the larger this region becomes. In rough terms, for given
values of the stress intensity factors, the size of this universal elastic region scales
as \ L, where L is the macroscopic scale on which forces are applied to the body.
Thus the assumptions of fracture mechanics become progressively better as samples
10
become larger and larger.
Outer elastic region: Far from the crack tip, stresses and strains are described by linear
elasticity. There is nothing more general to relate; details of the solution in this region
depend upon the locations and strengths of the loads, and the shape of the body. In
special cases, analytical solutions are available, while in general one can resort to nu-
merical solution. At first glance, the precise linear problem that must be solved might
seem inordinately complex. How can one avoid needing explicit knowledge of com-
plicated boundary conditions on some complicated loop running outside the outer
rim of the process zone? The answer is that so far as linear elasticity is concerned,
viewed on macroscopic scales the process zone shrinks to a point at the crack tip, and
the crack becomes a branch cut. Replacing the complicated domain in which linear
elasticity actually holds with an approximate one that needs no detailed knowledge
of the process zone is another approximation that becomes increasingly accurate as
the dimensions of the sample, hence the size of the universal elastic region, increase.
The dissipative processes within the process zone determine the fracture energy, # , defined
as the amount of energy required to form a unit area of fracture surface. In the simplest
case, where no dissipative processes other than the direct breaking of bonds take place, # is
a constant, depending on the bond energy. In the general case # may well be a complicated
function of both the crack velocity and history and differ by orders of magnitude from
the surface energy defined as the amount of energy required to sever a unit area of atomic
bonds. No general first principles description of the process zone exists, although numerous
models have been proposed (see e.g. [13]).
11
Mode I
Mode II
x
z Mode III
Figure 6: Illustration of the three conventional fracture modes, which are characterized by
the symmetry of the applied forces about the crack plane
cracks spontaneously orient themselves so as to make the Mode II component of the loading
vanish near the crack tip [17], as I will discuss in Section 2.11. Mode II fracture is however
observed in cases where material is strongly anisotropic. Both friction and earthquakes
along a predefined fault are examples of Mode II fracture where the binding across the
fracture interface is considerably weaker than the strength of the material that comprises
the bulk material. Pure Mode III fracture, although experimentally difficult to achieve, is
sometimes used as a model system for theoretical study since, in this case, the equations of
elasticity simplify considerably. Analytical solutions, obtained in this mode, have provided
considerable insight to the fracture process.
u _^
a`4 % ^ a `b (15)
2
12
where is analytic, and `c x % iy.
^
The asymptotic behavior of is easy to determine. Far from the hole, the stress
^
goes to a constant value dfe , where d is dimensionless. So
yz
ehg uy e2 i i^j x % iyY ! i^ x %
- iy Rk (16)
yz
j
l m`b,nog ! id for c ` nqp (17)
^j
How does the presence of the hole affect the stress field? Because the edges of the hole
are free, all stresses normal to the edge must vanish. Let t be a variable that parameterizes
the edge of the hole, so that
x t y t r
(18)
travels around the boundary of the hole as t moves along the real axis. Then
T ts g xt
g yt u and
N ts>! g yt
g xt u It is easy to verify that Nv wxTv =0, and because (19)
Tv is tangent Nv must be normal.
g g g g
are tangent vector and normal vectors along the edge of the hole, so requiring normal stress
to vanish means that
xz
yz Yy N 0 (20)
l ezs g ux
g uy u yBsB! g yt
g xt u 0 l g uy g xt ! g ux g yt 0 (21)
g g g g g g g g
l % x y
s !{g ix^ g ix^ u g t s g iy^ !|g iy^ u g t Insert (15) for u. (22)
g g g g g g
l g ^ t }g ^ t This equality holds only on the boundary. (23)
l g m`b+ g m`b Because ~ is arbitrary up to a constant any- (24)
^ ^ way, there is no worry about dropping a con-
stant of integration.
13
because 1 8 on the unit circle. Equation (27) is a relation between two analytic func-
tions that holds over the whole unit circle. The difference between f and 1 f is an
^ ^
analytic function; the Taylor series of the difference vanishes on the unit circle, and by an-
alytic continuation the difference vanishes everywhere, so the two functions must be equal
everywhere in the complex plane.
Now
f can be determined completely by analyzing its asymptotic behavior. Outside
^
of the hole must be completely regular, except for the fact that it diverges as ! i d` for
^
large ` . The relation between and ` is
` % ` 2 ! 4p
If the other sign of the square root were cho-
sen here, the argument below would have to
(28)
be modified in several places, but the final an-
2 swer would be the same.
Therefore, when ` is large, large, ` , and in accord with Eq. (17) one obtains
However,
f can have no other singularities within the unit circle, or else f would
^ ^
have corresponding singularities outside the unit circle, which is forbidden if u is to be
smooth away from the hole.
Having determined all the possible singularities of within and without the unit circle,
it is determined up to a constant. It must be given by
^
Displacements and stresses can now be determined from Eqs. (15) and (16). Figure 7 shows
a plot of yz along the line y 0 for a narrow ellipse with p 0 9.
2.5 Stress Intensity Factor
The limit p n 1 is particularly interesting. The elliptical hole closes down and becomes
a thin crack reaching from x ! 2 to x 2. The function acquires a branch cut over
^
14
e¡d e¡d e¡d
Radius of curvature R
e&d e¡d e¡d¢ yz
20
y
0
15
along y
x 10
yz ,
z 5
Stress
0 1 2 3 4
Distance x
in a solid are much greater than those
Figure 7: The stresses at the tips of an elliptical hole
applied off at infinity. The plot shows a solution of yz derived from Eq. (35) for d{ 1 and
m 0 9. Note the 20-fold increase in stress near the crack tip.
exactly the same region. The displacement u is finite everywhere, but the stress yz becomes
singular, diverging on the x axis as
n e¡d
-ehg uy
x e¡d
as x n 2
yz %
x ! 2\ x 2 \ x! 2
\ (35)
g
The stress intensity factor K is a coefficient of the diverging stress, defined by
r x
£ 0 \ 2¤ r yz
rlim K (36)
2 is the distance from the crack tip.
16
moving crack. I will not take it very far; just far enough to find the universal parts of the
fields in the vicinity of the crack tip.
The starting point is the equation of motion for an isotropic elastic body in the contin-
uum limit,
g t 2 °x± % e¡ ] ] y6u % e ] 2u
2u
(40)
g
originally found by Navier. u is a field describing the displacement of each mass point
from its original location in an unstrained body, is the density, and the constants e and ±
are called the Lamé constants, have dimensions of energy per volume, and are typically of
order 1010 ergs/cm3 .
The tensor structure of elasticity makes it particularly difficult to solve fully three-
dimensional problems, and it is difficult to carry out controlled three-dimensional exper-
iments as well. Fortunately, there are cases in which the theory naturally reduces to two
dimensions, where most of the analytical results have been obtained.
A first case is called anti-plane shear. Imagine one is tearing a telephone book, with
one hand gripping on the left, the other gripping on the right, one pushing up and the other
pulling down. The only non-zero displacement is uz , and it is a function of x and y alone.
The only non-vanishing stresses in this case are
uz
xz ²e g x
(41)
and g
uz
yz ²eg y
(42)
c
e (44)
Therefore, the vertical displacement obeys the ordinary wave equation.
A second case corresponds to pulling on a thin plate, and is called plane stress. Let
the z direction be the direction that goes through the plate. If the scale over which stresses
are varying in x and y is large compared with the thickness of the plate, then one might
expect that the displacements in the z direction will come quickly into equilibrium with the
local x and y stresses. When the material is being stretched, (think of pulling on a balloon),
the plate will contract in the z direction, and when it is being compressed, the plate will
thicken. Therefore, one guesses that
uz z f u x uy
(45)
and that ux and uy are independent of z. One can deduce the function f by noticing that
must vanish on the face of the plate. This means that
zz
f ux uy &g
uz
z
'! ± % ± 2e ³ g uxx % g uyy ´ (47)
g g g
So
g ux
x
% g uy
y
% g uz
z
± % 2e 2e ³ g uxx % g uyy ´
(48)
g g g g g
and one can write
¦=µ ± ¶ I¦ µ g ux·· % ezs g uxµ¦ % g x¦µ u
u
(49)
g g g
where
±¸ ¶ ± 2% e¡2± e
(50)
and ¨ and ¹ now range only over x and y. Therefore, a thin plate obeys the equations of
two-dimensional elasticity, with an effective constant ± ¶ , so long as uz is dependent upon
ux and uy according to Eq. (47). In the following discussion, the tilde over ± ¶ will usually
be dropped, with the understanding that the relation to three-dimensional materials prop-
erties is given by Eq. (49). The equation of motion is still Navier’s equation, Eq. (40), but
restricted to two dimensions.
A few random useful facts: materials are frequently described by the Young’s modulus
E and Poisson ratio . In terms of these constants,
The following relation will be useful in discussing two-dimensional static problems. First
note that
] y u ¼x± % 2e¡ ¦=¦ (52)
Second, taking the divergence of Eq. (40), one finds that
2
¦=¦ ]
± % 2e g 2 ¦E¦
2
(53)
t
] g
Therfore, yºu obeys the wave equation, with the longitudinal wave speed
%
cl
± 2e (54)
Similarly,
]½ u also obeys the wave equation, but with the shear wave speed
ct
e (55)
The fact that the local thickness of the plate is tied to stresses in the x and y directions
leads to two optical methods to determine stress fields experimentally. The first method re-
lies on the fact that when light reflects off a curved surface, the reflected intensity becomes
18
Figure 8: Interference fringes captured during rapid motion of crack though brittle plastic.
K Ravi–Chandar.
singular at certain points that depend on the details of the geometry. In practice, when light
is shined on a crack tip, a sharp dark spot surrounds the tip, and its shape and size can be
used to deduce the stresses. This technique is known as the method of caustics. A second
method, older and more reliable, relies upon the fact that materials under stress will typ-
ically rotate the plane of polarization of transmitted light. It is possible to determine the
basic structure of this rotation without detailed calculation. The rotation must depend upon
features of the stress tensor which are rotationally invariant, and therefore can depend only
upon the two principal stresses, which are the the diagonal elements in a reference frame
where the stress tensor is diagonal. In addition, there should be no rotation of polarization
when the material is stretched uniformly in all directions, in which case the two principal
stresses are equal. So the angular rotation of the plane of polarization must be of the form
K 1 ! 2
(56)
where 1 and 2 are the principal stresses (eigenvalues of the stress tensor), and K is a
constant that will have to be determined experimentally. Whenever stresses of a two-
dimensional problem are calculated analytically, the results can be placed into Eq. (56),
and compared with experimental fringe patterns. Fast optical systems have been developed
to carry out this procedure for rapidly moving cracks. Although one might wonder about
the extent to which Eq. (56) is obeyed when cracks move at speeds on the order of the
speed of sound, experimental results match rather well with predictions for crack tip fields.
¨ 2 1!
¬ 2
± % 2e (63a)
¹ 2 1 ! ¬e
2
(63b)
t ² t0 z % t0 z % t1 x % i¹ y % t1 x % i¹ y
(65)
subject to the constraint of Eq. (60), which gives a relation between l0 and t0 .
In fact, the purely harmonic pieces l0 and t0 disappear entirely from the expressions
for u They result from the freedom one has to add a harmonic function to l and t simul-
taneously, and can be neglected. Defining z ¡ l1 z2 g z and Ác z¯ g t1 z2 g z I have
for u
^ g
ux z¦ % z¦Y % i¹{ÂÃÁÄ zµBÅ ! ÁÄ zµ4ÅIÆ (66a)
^ ^
uy i ¨ÈÇ z¦Y! z¦É&!|ÂPÁÄ zµ4Å % ÁÄ zµBÅÊÆ (66b)
^ ^
where
z¦U x % i ¨ y zµ x % i¹ y (67)
Equation Eq. (66) gives a general form for elastic problems which are in steady state
moving at velocity . Define also Ë z2 g z and ÌÍ g ÁÎ z r g z Then the stresses are
given by
g ^
% % ¡e ¡ÇPË¿ z¦ % Ë¿ z¦YÏÉÐÄ 1 !®¨ 2 Å
2 x± (68a)
xx yy
20
we will also need the rotation, which means
g uy
x
!Ög ux
y
"!×Ä 1 !¿¹ 2 Å ÂKÌ²Ä zµ4Å % Ì²Ä zµ4Å=Æ (69)
g g
It is worth writing down the stresses directly as well:
yy '!e Ä 1 % ¹ 2 Å Ç Ë¿ z¦Y % Ëz z¦ É ! 2i¹¡e Â Ì Ä zµ4Å ! Ì Ä zµ4ÅIÆ (70)
²eØÄ 1 % 2¨ 2 ¿
xx ! ¹ 2 Å ÇÙËz z¦Ó % Ë¿ z¦ÏÉ % 2i¹¡eÈÂPÌ²Ä zµ4Å ! Ì²Ä zµ4Å=Æ (71)
The definitions of ¨ and ¹ in Eq. (63) have been used to simplify the expressions.
To solve a general problem, one has to find the functions and Á which match boundary
^
conditions. It is interesting to notice that when ¸n 0, the right hand side of Eq. (68a)
goes to zero as well. Since one will be finding the potentials from given stresses at the
boundaries, Ë must diverge as 1 I , and the right hand side of Eq 68 will turn into a
derivative of Ë with respect to ¨ . That is why the static theory has a different structure than
the dynamic theory. In fact, the dynamic theory is more straightforward.
As a first application, I will show that a moving crack under symmetric loading becomes
unstable at a certain speed. Assume the crack to lie along the negative x axis, terminating
at x 0 and moving forward. The problem is assumed symmetric under reflection about
the x axis, but no other assumption is needed. This instability was first found in a particular
case by Yoffe (1951).
We know that in the static case, the stress fields have a square root singularity at the
crack tip. I will assume the same to be true in this case (the assumption is verified in all
cases that can be worked explicitly.) Near the crack tip, I assume that
% iBi z1 Ú
^ z Br
2
(72)
Ác z Dr % iDi z1 Ú 2
(73)
We first appeal to symmetry. Observe that
We also observe that the square roots in Eq. (73) must be interpreted as having their cuts
along the negative x axis, corresponding to the crack.
On the crack surface, we have two
boundary conditions, which require that xy and yy vanish. Upon substituting Eq. (75)
into Eq. (71) we find that the condition upon yy is satisfied identically for x Ü 0 y 0.
However, substituting into Eq. (69) with y 0 we find that
²e i Ý 2¨ Br ! Ä ¹ 2 % 1Å Di Þ ³ 1
! 1
(76)
\ x´
xy
x \
21
Thus
Di 2¨
Br ¹ % 1
2
(77)
This relation is enough to find the maximum velocity at which a crack can proceed stably
along the x axis.
Using Eq. (77) we find that
K
Ä ¹ 2 % 1Å Ä 1 % 2¨ 2 !¿¹ 2
Å ³ 1
!%4ß
¨ ¹ ³ 1 %
1 1
k (78a)
\ 2¤ D i z¦
\ ´ z ¦
\ zµ \ z µ ´
xx
\
K
!i Ä 1 % ¹ 2 Å ³ 1z¦ % 1 % 4ߨ ¹ ³ 1 % 1 k
2
(78b)
yy
\ 2¤ D \ \ z ¦Î´ \ zµ \ z µ ´
2i ¨ Ä ¹ 2 % 1Å ³
K 1 1 1 % 1
! ! (78c)
xy
\ 2¤ D \ z¦ \ z ¦ \ zµ \ z µ ´
with
D 4¨ß¹$!à 1 % ¹ 2 2 (78d)
The constant K is again called the stress intensity factor, and is given by
K lim x n 0á \ 2¤ x (79)
yy
In order to find the direction of maximum stress, one must approach the tip of the crack
along a line at angle to the x axis, and compute the stress perpendicular to that line. So
one wants to choose
z¦U r cos % %
ri ¨ sin
zµ r cos ri¹ sin
(80)
and to evaluate the stress
2 cos2 yy % sin2 xx ! sin 2E xy (81)
When this is carried out, one finds that above a certain velocity (for Poisson ratio â 1 3,
the critical velocity is about .61) the direction of maximum tearing stress points away from
the axis, and the crack would presumably become unstable. The point at which this happens
is referred to as the Yoffe instability.
The factor 4 ¨ß¹$!à 1 % ¹ 2 2 is a function of crack velocity whose roots happen to de-
fine the Rayleigh wave speed, which is the speed at which waves travel over a free surface.
Because this factor vanishes at the Rayleigh wave speed and becomes negative for larger
values of crack velocity, the Rayleigh wave speed is conventionally thought to provide an
upper limiting speed for cracks. The argument seems fairly secure for this loading geome-
try. However, cracks loaded in Mode II, or moving along interfaces can go faster [21, 22]. I
think Petersan, Deegan, and I have some experimental evidence for cracks traveling faster
than the shear wave speed under Mode I loading in rubber, but we are not yet 100% certain.
23
zone; Hodgdon and Sethna [24] show how to arrive at this conclusion using little more than
symmetry principles.
Cotterell and Rice also demonstrated that the condition KII 0 has the following con-
sequences for crack motion. Consider an initially straight crack, propagating along the x
axis. The stress field components xx and yy have the following form
KI % %àé r1Ú 2
2¤ r 1 Ú
xx T (89a)
2
KI %àé r1Ú 2
2¤ r 1 Ú
(89b)
yy 2
The constant stress T is parallel to the crack at its tip. Cotterell and Rice showed that if
T ê 0, any small fluctuations from straightness cause the the crack to diverge from the x
direction, while if T Ü 0 the crack is stable and continues to propagate along the x axis.
They also discuss experimental verification of this prediction.
Yuse and Sano [25] and Ronsin, Heslot, and Perrin [26] have conducted experiments
by slowly pulling a glass plate from a hot region to a cold one across a constant thermal
gradient. The velocity of the crack, driven by the stresses induced by the nonuniform
thermal expansion of the material, follows that of the glass plate. At a critical pulling
velocity, the crack’s path deviates from straight-line propagation and transverse oscillations
develop. This instability is completely consistent with the principle of local symmetry
[27,28]; the crack deviates from straightness when T in Eq. (89) rises above 0. Adda-Bedia
and Pomeau and Ben-Amar [29] have extended the analysis to calculate the wavelength of
the ensuing oscillations.
Hodgdon and Sethna [24] have generalized the principle of local symmetry to three
dimensions. They show that an equation of motion for a crack line involves in principle
nine different constants. It would be interesting for an experimental study to follow upon
their work and try to measure the many constants they have described, but we are not aware
of such experiments. Larralde and Ball [30, 31] analyzed what such equations would imply
for a corrugated crack, and found that the corrugations should decay exponentially. They
also performed some simple experiments and verified the predictions.
Thus the principle of local symmetry is consistent with all experimental tests that have
been performed so far on slowly moving cracks. Nevertheless, it does not rest upon a
particularly solid foundation. There is no basic principle from which it follows that a crack
must extend perpendicular to the maximum tensile stress, or that it must maximize energy
release.
The logic of the principle of local symmetry says that bonds under the greatest tension
break first, and therefore cracks loaded in Mode I move straight ahead, at least until the
velocity identified by Yoffe when a crack is predicted to spontaneously break the symmetry
inherent in straight-line propagation. This logic is called into question by a very simple
calculation, first described by Rice
[32].
Let us look at the ratio xx yy right on the crack line. From Eq. (78) it is
2% %
a¹ 14º ¯¨ 1¹$!à2 m1¨ % !¿¹ ¹2 2! 4¯¨ ¹
2 2
xx
(90)
yy
24
2 §¹ 2 % 1 ;m¨ 2 !¿¹ 2 !
4¨ß¹$!à 1 % ¹ 2 2 1 (91)
1%
2 ct4 % c4l %
2c2l ct2 cl ! ct º cl % ct
(92)
r22
and in fact Eq. (91) greater than unity for all . This result is surprising because it states
that, in fact, the greatest tensile forces are perpendicular to a crack tip and not parallel to it,
as soon as the crack begins to move. Therefore it is hard to understand why a crack is ever
supposed to move in a straight line.
These remarks do not do justice to the calculations performed with the cohesive zone
models. In their most elaborate versions, the crack is allowed to pursue an oscillating path,
and the cohesive zone contains both tensile and shear components. In most, although not
all of these models, crack propagation is violently unstable to very short-length oscillations
of the tip.
A summary of this work has been provided by Langer and Lobkovsky [33]. They
consider a large class of models, and correct subtle errors in previous analyses. They do
find some models in which a crack undergoes a Hopf bifurcation to an oscillation at a
critical velocity. However, their “general conclusion is that these cohesive-zone models
are inherently unsatisfactory for use in dynamical studies. They are extremely difficult
mathematically and they seem to be highly sensitive to details that ought to be physically
unimportant.” Attempts to find models in this general class continue, with recent attempts
by Karma, Kessler and Levine [34] and Aranson, Kalatsky Vinokur [35]. Both of these
papers model Mode III cracks, and it is not clear how to generalize them to Mode I. The
more recent paper cures some problematic features of the first one; I am not convinced that
either provides a really useful starting point to address outstanding problems in fracture.
One possibility is that cohesive zone models must be replaced by models in which
plastic yielding is distributed across an area, and not restricted to a line. Dynamic elastic-
plastic fracture has not, to our knowledge, considered cracks moving away from straight
paths. Another possibility is that these calculations signal a failure of continuum theory, and
that the resolution must be sought at the atomic or molecular scale. It is not possible right
now to decide conclusively between these two possibilities. However, the experimental
observation that the dynamic instabilities consist of repeated frustrated branching seems
difficult to capture in a continuum description of the process zone.
3 Incorporating Plasticity
The lectures included an improvised segment on elastic–plastic fracture mechanics. There
was no written material prepared, but here is a brief outline:
Elastic–plastic fracture mechanics is a large and well–developed subject within engi-
neering, resting on a large scientific base that considers interactions of cracks and disloca-
tions.
Thomson Physics of Fracture (1986) ; Weertman Dislocation–Based Fracture Mechan-
ics (1996) ; Janssen, Zuidema, and Wanhill Fracture Mechanics (2002)
25
Basic results:
When plastic deformation at tip is significant, cracks move slowly, and bond–breaking
is irrelevant.
Total plastic dissipation can be estimated from irreversible separation of crack faces
(Crack Tip Opening Displacement):
#) c
(93)
Steady states: Steady state crack motion exists, and is a stable attractor for a range of
energy flux.
26
Phonons: Steadily moving cracks emit phonons whose frequencies can be computed from
a simple conservation law.
Fracture energy The relation between the fracture energy and velocity can be computed.
Velocity gap: The slowest steady state runs at around 20% of the Rayleigh wave speed;
no slower-moving steady state crack exists.
Instability: At an upper critical energy flux, steady state cracks become unstable, and
generate frustrated branching events in a fashion reminiscent of experiments in amor-
phous materials.
One may wonder about the motivation for formulating a theory for the dynamic behav-
ior of a crack in a crystal. Is the lattice essential or can one make the underlying lattice go
away by taking a continuum limit? So far as I know, all attempts to describe the process
zone of brittle materials in a continuum framework have run into severe difficulties [33].
These problems do not arise in an atomic-scale description.
The simplicity of the ideal brittle crystal is somewhat misleading in a number of re-
spects. Therefore, before introducing the mathematics, I will comment on a number of
natural questions regarding the generality of the predictions it makes.
Does the simple force law employed between atoms neglect essential aspects of the
dynamics? I will demonstrate that the same qualitative results observed in the ideal
brittle crystal occur in extensive molecular dynamic simulations using realistic po-
tentials.
The calculations are in a strip. Is this geometry too restrictive? To this question
the general formalism of fracture mechanics provides an answer. Fracture mechanics
tells us that as long as the conditions of small-scale yielding are satisfied, the behavior
of a crack is entirely governed by the structure of the stress fields in the near vicinity
of the crack tip. These fields are solely determined by the flux of energy to the crack
tip. A given energy flux can be provided by an infinite number of different loading
configurations, but the resultant dynamics of the crack will be the same. As a result,
the specific geometry used to load the system is irrelevant and no generality is lost by
the use of a specific loading configuration. This fact is borne out by the experimental
work described in the previous section.
Have lattice trapping and the velocity gap ever been seen experimentally? Maybe.
Robert Deegan and I have obtained some results, but the data are not yet conclu-
sive. Molecular dynamics simulations indicate that lattice trapping disappears at
room temperature. New experiments are needed to obtain a detailed description of
dynamic fracture in crystals at low temperatures.
Are results in a crystal relevant for amorphous materials? This is still an open ques-
tion. However, the results of the lattice calculations seem to be remarkably robust.
Adding quenched noise to the crystal has little qualitative effect. Effects of topolog-
ical disorder are not known. However, a certain amount is known about the effects
27
of increasing temperature. When the temperature of a brittle crystal increases above
zero, the velocity gap closes [41], and its behavior is reminiscent of that seen in
experiments performed on amorphous materials.
Kulakhmetova, Saraikin and Slepyan [9, 42] first showed that it is possible to find exact
analytical expressions for Mode I cracks moving in a square lattice. They found exact rela-
tionships between the energy flux to a crack tip and crack tip velocity. They also observed
that phonons must be emitted by moving cracks, and calculated their frequencies and am-
plitudes. Later calculations extended these results to other crystal geometries, allowed for a
general Poisson ratio, showed that there is a minimum allowed crack velocity, found when
steady crack motion is linearly stable, calculated the point at which steady motion becomes
unstable to a branching instability, and estimated the spacing between branches [38].
Figure 9: A sketch showing steady state motion of crack moving in an ideal brittle crys-
talline strip loaded in Mode I.
These calculations are extremely elaborate. The analytical expressions are too lengthy
to fit easily on printed pages, and most of the results were obtained with the aid of symbolic
algebra. For this reason, I will first give a qualitative argument for the most surprising of
the dynamical results, the velocity gap. Then, after summarizing the results of the technical
calculations, I will proceed to describe in detail how the calculations work in the case of a
simple one–dimensional model where the algebra is much less demanding but most of the
ideas are the same.
28
crystals. In the analytically solvable models, the bond breaking is instantaneous. In any
more realistic situation there will instead be a brief period in which the force between atoms
falls to zero, but all that is required for the velocity gap is that this time be short compared
to the vibrational period of an atom.
Dynamic fracture is a cascade of bonds breaking, one giving way after another like a
toppling line of dominos. Examining Figure 10, watch what happens as a crack moves
ahead. In frame 2, the bond between two atoms has just broken. There is no guarantee
that the next bond to the right will break. The crack could fall into a static lattice trapped
state. The best chance to avoid this fate is for the atom marked in green to deliver enough
of a blow to its right-hand neighbor that the bond on that neighbor is broken in turn. This
process had better take place within the first half of the first vibrational period of the green
atom. For the longer it vibrates, the more energy will be distributed to its neighbors in all
directions in the form of traveling waves, and the smaller become the chances that there
will be enough concentrated energy available to snap the next bond down the line.
To make this idea slightly more quantitative, let ì b be the effective spring constant
acting on an atom after a bond connected to it has snapped. Its oscillatory period is
í 2¤
M
(95)
ì b
where M is its mass. Assuming that the next bond, at distance a, breaks during some
fraction ¨Ü 1 of half this oscillatory period, the speed of the crack will be
¨¡¤a ì b
(96)
M
By this same rough logic, the speed of sound is given by solving the wave equation
2u
M uå ²ì a 2
g (97)
x2
g
where now ì is an effective spring constant appropriate for atoms surrounded by unbroken
bonds. From Eq. (97) follows a wave speed c of
c a
ì (98)
M
One should expect c to be larger than the crack speed . The main reason is that when
one of the springs connected to an atom is snapped, the effective spring constant describing
its vibrations should decrease, so ìîê-ì b . There is also an extra factor of ¤ in Eq. (96).
The crack speed estimated here is a lower bound because there is no way that pulling
more gently on remote system boundaries can reduce the vibrational period of two atoms
once the bond connecting them breaks. Pulling harder on external boundaries can however
increase the speed of a crack because it can supply enough energy that the atoms are already
moving quickly when the bond between them snaps.
These arguments presume that the only source of energy available for snapping bonds
is contained in strained material ahead of the tip. This assumption is never completely
29
1 5
2 6
Snaps Snaps
3 7
4 8
Figure 10: Sequence of eight snapshots from motion of a crack tip. Bonds break in frames
two and six. The green atom snaps backwards each time a bond connected to it breaks,
providing a crucial portion of the energy for the next bond to snap. This process must
occur within the first half of the first vibrational period of the green atom, or it is unlikely
ever to occur.
30
correct. At any nonzero temperature, thermal fluctuations can bring extra energy in. For
this reason, cracks strained above c always move ahead at any nonzero temperature, but the
rate is very small when the thermal fluctuations are rare. Similarly, chemical agents in the
environment can help catalyze bond breaking, and crack speeds are sometimes controlled
by the rate at which they arrive. Dynamic fracture refers to the motion of cracks when these
sorts of effects can be neglected.
dè 2 ð
dè 1 ð0ñ
dè 1 ï dè 2 ï dè 6 ð
dè 5 ï dè
dè 5 ð èd3 ï dè 4 ï 6 ï
dè 4 ð
dè 3 ð
u 2
u 1
u 0 u 6
u 5
u 3
u 4
Figure 11: The geometry of the lattice used for fracture calculations.
31
÷¬ø
ù ÷¬ø
õö ú ù ÷ø
xm ü
1 N
1
ú ù
1 N
õö ú
1
1 N
xm
1
xm û 1
xm û 1
õö ÷Eø
xm ù
õö ÷ø
1
÷ø ù ú
1 N
xm ü 1
ù ú
1
1 N
ú
1 N
ý þ ÿ
2N 1
Figure 12: In this simplest of all solvable atomic crack models, two lines of atoms are
connected by weak springs to a ceiling and a floor, and to horizontal and vertical neighbors
by strong springs. The vertical strong springs break if they exceed a certain extension.
32
linear springs until the instant they snap at a separation of 2u f , from which point on they
exert no force. The location of each mass point is described by a single spatial coordinate
u m n , which can be interpreted as the height of mass point m n into or out of the page.
The force between adjacent masses is determined by the difference in height between them.
The index m takes integer values, while n takes values of the form 1 2, 3 2, N % 1 2.
The model is described by the equation
2r
Figure 13: Dynamic fracture of a triangular crystal in anti-plane shear. The crystal has
2 N % 1 rows of atoms, with N 4, and 2N % 1 rows of bonds. Heights of spheres indicate
displacements u m n of mass points out of the page once displacements are imposed at the
boundaries. The top line of spheres is displaced out of the page by amount u m N % 1 2
UN \ 2N % 1, and the bottom line into the page by amount u m
! N ! 1 2
¯¼! UN
! \ 2N % 1. Lines connecting mass points indicate whether the displacement between
them has exceeded the critical value of 2u f ; see Eq. (99b)).
©
uå m n Y! bu %
1
f u m n ! u m n Rk (99a)
ä 2 nearest
P¥ i j
j
neighbors m n
with
f u & u > 2u f ! u (99b)
describing ideal brittle springs, the step function, and b the coefficient of a small dissipa-
tive term. There is no difficulty involved in choosing alternative forms of the dissipation, if
desired. The boundary condition which drives the motion of the crack is
u m N % 1 2k>
Øi
UN (99c)
It is important to find the value of UN for which there is just enough energy stored per
length to the right of the crack to snap the pair of bonds connected to each lattice site on
33
the crack line. For m 0 one has
nUN
u m n &
N % 1 2
(100)
and the height difference between mass points with adjacent values of n is
UN
Uright N % 1 2
(101)
Therefore, the energy stored per unit length in the 2N % 1 rows of bonds is
1½ ½ ½ ½
2 Upper Bonds Sitek Rowsk Spring Constantk Uright
2
(102)
2 i i i
2 2N % 1 %
1 1 UN
2 2 N 1 2
2 (103)
2Q0 UN 2
(104)
with
Q0 2N1% 1 (105)
The energy required to snap two bonds each time the crack advances by a unit length is
1 ½ 2 Bonds Sitek
½ Spring Constantk
½
i Separation at fracturek
2
(106)
2 i i
1 ½ 2
½ 1 ½ 2u 2 2u2f (107)
2 2
f
Therefore, equating Eqs. (104) and (107) the proper dimensionless measure of the external
driving is
UNu\ Q0
(108)
f
a quantity which reaches 1 as soon as there is enough energy to the right of the crack to snap
the bonds along the crack line, and which is linearly related to the displacements imposed
at the edges of the strip.
4.4 Symmetries:
Assume that a crack moves in steady state, so that one by one, the bonds connecting
u m 1 2 with u m % 1 ! 1 2 or u m ! 1 2 break. They break because the distance
between these points exceeds the limit set in Eq. (99b) and as a consequence of the driving
force described by Eq. (99c). Assuming that the times at which bonds break are known,
the original nonlinear problem is immediately transformed into a linear problem. However,
one has to come back at the end of the calculation to verify that
1. Bonds break at the time they are supposed to. Imposing this condition determines a
relation between crack velocity and loading .
34
2. No bonds break when they are not supposed to. Imposing this condition leads to the
velocity gap on the low-velocity end, and to crack tip instabilities above a critical
energy flux.
Steady states in a crystal are more complicated than steady states in a continuum. In a
continuum, steady state acts as u x !z t . The closest one can come in a triangular crystal
is by having the symmetry
We have defined
0 if n 1 2 5 2
r2
gn 1 if n 3 2 7 2
r2
(110)
mod n ! 1 2 2 in general
4.5 Eliminating the spatial index m:
Assuming that a crack is in steady state, we can therefore eliminate the variable m entirely
from the equation of motion, by defining
un t , u 0 n t (111)
and write the equations of motion in steady state as
% un 1 t !à gn ! 12 ÊB % u n 1 t ! gn I4
uå n t B
1
%% á
un t % 1 I4
á 1
! 6un t % á
un t ! 1 I4
á 1
c! bun (112a)
2
un 1 t !à gn 1 ! 12 ÊB %
u n 1 t ! gn 1 I4 ä
if n ê 1 2, and
%
u3 Ú t t ! 1 Ê B % u3 Ú
% % 2
t 1I4! 4u t % t ! 1 ÊB
2
% 21Ú 2 t ! u1Ú 2 t R1kN>Ú 2R! t % 21Ú 2 t ! 1I4Y!
1 u1 Ú u1 Ú
uå 1 Ú t B (112b)
u1 Ú 2 t RkN> 1 > 2 4Y! t
2 u u
2
i i ! buä 1Ú 2
if n 1 2
The time at which the bond between u 0 1 2 t and u 0 ! 1 2 t breaks has been cho-
sen to be t 0, so that by symmetry the time the bond between u 0 1 2 t and u 1 ! 1 2 t
breaks is 1 2 .
35
n 1 2. Fourier transforming Eq. (112a) in time gives
Ú k
i gn 1 1
Ú %
% 1
2 un 1á
f i eei>Ú ! 6 % e i>eÚ
i
k
gn 1
h! un f, ib %
2
f
1 (113)
2 un
% 1
2 un 1
f
Ú k
i ei gn 1 1
Ú % ei gn 1
i
f ek n 1Ú 2 !Ú !
Let
un f& u1 Ú i gn 2
(114)
2
Substituting Eq. (114) into Eq. (113), and noticing that gn % gn
á 1 1 gives
k ei gn 1 gn 2
Ú" 2! % ei gn 1 gn
Ú! 2! k
2 1Ú 2
1
u
f
e á á
!@ 2u1Ú 2 f, ib u1Ú 2
f % % 12 u1Ú 2 f ii ei>Ú ! 6 % e i>Ú k
% 1 u1Ú 2 f e k ei gn 1 á gn 2
"Ú 2! % ei# gn 1 á gn
Ú" 2! k
2 i (115)
l 2 % ib % 2 cosh k cos
f> 242 % cos
fI4Y! 3 0 (116)
Defining
3 ! cos
fI4Y! 2 ! ib
z (117)
2 cos
f 2B
one has equivalently that
y z%
z2 ! 1 (118)
with
y ek (119)
One can construct a solution which meets all the boundary conditions by writing
y$ á Ú % ! y $ á Ú %
f e Ú k% UN n ! 1 2 2¨
N 1 2 n N 1 2 n
y ! y
un f& u1 Ú i gn 2
2
i ¨ % 2 (120)
N N N 2
now gives
u1 Ú f F f!à 1 % ei>Ú U 2
f,"! UN 2¨
% ¨ 2
(125)
2
N 2
$ % ! y $ % !
with
yN 1 N 1
%
y ! y
F
f& ³ 2z cos f 24 1 (126)
N N ´
Next, use Eq. (122) in the form
1% e >Ú
i 2
U f, u1 Ú 2 f (127)
2
to obtain
U f F
fY! 2 cos2 f 4« U
f&"! UN 2¨
N 2%
(128)
¨ 2
Writing
U
f+ U áf
f % U
f (129)
finally gives
U áf
f Q f %
f, UN Q0 Ñ
1 % 1
U %
¨ i ¸ ¨ ! i Ô
(130)
with
F
f
Q
f+ (131)
F
fY! 2 cos2 f 40
To obtain the right hand side of Eq. (130) one uses the facts that F 0 &"! 1 N, and that ¨
is very small, so that the right hand side of Eq. (130) is a delta function.
The Wiener-Hopf technique [46] directs one to write
Q
f
Q f& (132)
Qá
f
where Q is free of poles and zeroes in the lower complex plane and Q á is free of poles
and zeroes in the upper complex plane. One can carry out this decomposition with the
explicit formula
d
& j ln Q j
Q f> exp lim
i' £ 0 ã 2 ¤ i ! i )(+* j
k (133)
Now separate Eq. (130) into two pieces, one of which has poles only in the lower half
plane, and one of which has poles only in the upper half plane:
U á
f Q0UN 1
Q áhf
!
Q 0 J! i % ¨Ð
Q0UN 1 U
f
Q 0 i % !
¨æ Q f (134)
37
Because the right and left hand sides of this equation have poles in opposite sections of the
complex plane, they must separately equal a constant, . The constant must vanish, or U ,
and U á will behave as a delta function near t 0. So
U
Q0 Q f
f, UN
Q 0 ;a¨ % if
(135a)
and
Q0 Q áf
Q
U áf, UN (135b)
0ºm¨¸! if
One now has an explicit solution for U
f . Numerical evaluation of Eq. (133), and U t
from Eq. (135) is fairly straightforward, using fast Fourier transforms. However, in carrying
out the numerical transforms, it is important to analyze the behavior of the functions for
large values of . In cases where functions to be transformed decay as 1 i , this behavior
is best subtracted off before the numerical transform is performed, with the appropriate
step function added back analytically afterwards. Conversely, in cases where functions to
be transformed have a step function discontinuity, it is best to subtract off the appropriate
multiple of e t > t before the transform, adding on the appropriate multiple of 1 > 1 ! if
afterwards. A solution of Eq. (135) constructed in this manner appears in Figure 15.
lim
d
e Ui t
f, uf (137)
t £ 0 ã 2¤
This integral can be evaluated by inspection. One knows that for positive t ê 0,
d exp
i ! i t k U f, 0
ã (138)
and that any function whose behavior for large is 1 i has a step function discontinuity
at the origin. Therefore, Eq. (137) and Eq. (135a) become
Q mp
uf UN Q0
Q 0 (139)
Since from Eq. (131) it follows that Q xp & 1, one sees from Eq. (133) that
38
Q xp & Qá xp & 1 (140)
As a result, one has from Eq. (139) and the definition of given in Eq. (108) that
Q 0
(141)
\ Q0
To make this result more explicit, use Eq. (133) and the fact that Q R!@f0 Q f to
write
¾
Q
d 1 ln Q
% ln Q J!h
0> exp j
2¤ 2 i Ó! i
j * % i * j kÀ (142)
¾ã j j
d
j
1
ln ³
Q %
j
* ln Q 0 Ïk À
exp
ã 2¤ i ! 2i
¾
Q ´
j j * 2 %
j
2
l Q 0> Q0 exp ! ã d
j
1
ln ³
Q
j À (143)
2¤ 2i Q
j j ´
Placing Eq. (143) into Eq. (141) gives
d 1 Q
exp
i! ã j
2¤ 2i
ln ³
Q
j ´ k (144)
j j
In order to record a final expression that is correct not only for the Mode III model
considered here, but for more general cases, rewrite Eq. (144) as
d
j 1
C exp
i! ã 2¤ 2i
j
Ý ln Q
j ! ln Q k
j Þ
(145)
where C is a constant of order unity that is determined by the geometry of the lattice,
equaling 1 for the triangular lattice loaded in Mode III, but 2 \ 3 for a triangular lattice
loaded in Mode I [38]. When written in this form, Eq. (145) is suitable for numerical
evaluation, since there is no uncertainty relating to the phase of the logarithm.
When b becomes sufficiently small, Q is real for real except in the small neighborhood
of isolated roots and poles that sit near the real axis. Let riá be the roots of Q with negative
imaginary part (since they belong with Q á ), ri the roots of Q with positive imaginary part,
&
and similarly pi the poles of Q. Then one can rewrite Eq. (145) as
1C /00 2 r pá i i
for b n 0 (146)
rá p i i
One may derive Eq. (146) as follows: away from a root or pole of Q, the integrand of
Eq. (145) vanishes. Consider the neighborhood of a root r á of Q which falls to the real
axis from the negative side as b n 0. For the sake of argument, take the imaginary part of
this root to be ! ib. In the neighborhood of this root, say within a distance \ b, the integral
to compute for Eq. (145) is
1 r á43 b d
¾
ln !à r á ! ib2
! 3 j j
! ln
!à r á % ibr À
(147)
2¤ ãr b 2i
j j
39
Defining ! r á , and integrating by parts gives
jj j
! 21¤ ã 3 bd % 2ib
3 b 2i
j j ln i j j rá k 2 % b2
(148)
1 %àé \ jj
-! ln r á b (149)
2
Similar integrals over other roots and poles of Q finally produce Eq. (146).
Together with Eq. (135), Eq. (145) and Eq. (146) constitute the formal solution of the
model. Since Q is a function of the steady state velocity , Eq. (144) relates the external
driving force on the system, , to the velocity of the crack . The results of a calculation
appear in Figure 14.
0.8
Steady states go unstable
0.6
0.2
Velocity gap
0.0
Linearly stable lattice-trapped states
1.0 1.1 1.2 1.3 1.4
Figure 14: The velocity of a crack 4 c, scaled by the sound speed c \ 3 2, is plotted as
a function of the driving force . The calculation is carried out using Eq. (146) for N 9.
40
5.0
4.0
u(t) 3.0
2.0
1.0
0.0
-50 -25 0 25 50
t
Figure 15: A plot of U t for Î 5, N 9, and b 0 01, produced by direct evaluation of
Eq. (135). Note that mass points are nearly motionless until just before the crack arrives,
and that they oscillate afterwards for a time on the order of 1 b.
3.0 798 :
5
2.5
798 :
2
@ 2.0
k
?= > 1.5
<;
k
1.0
0.5
0.0
0 2 4 6 8 10
k
Figure 16: Graphical solution of Eq. (153), showing that for low velocities, a large number
of resonances may be excited by a moving crack.
41
the phonons are described by the dispersion
¯¦¡ k (150)
If the particle moves with constant velocity , and interacts with the various ions according
A
to some function , then to linear order the motions of the ions can be described by a matrix
D which describes their interactions with each other as
©C
C C
B B R l ! R l ul % A B R !¿ t
©
muå l "! D l
(151)
l l
DFE
Multiplying everywhere by eik R , summing over l, letting K be reciprocal lattice vectors,
l
G
and letting be the volume of a unit cell gives C C
B ©C B
muå k Y D k u k %
HD E B
©
G E eE á A k % K
1 i k K t
(152)
K
Inspection shows that the lattice frequencies excited in this way are those which in the
extended zone scheme[ [47]] obey
Pretending that the crack is a particle, one can use Eq. (153) to predict the phonons that the
crack emits.
There is another version of this argument that is both simpler and more general. The
only way to transport radiation far from a crack tip is in traveling waves. However, in
steady state, the traveling waves must obey symmetry (109a). In a general crystal with
lattice vectors R and primitive vectors a , applying this requirement to a traveling wave
exp ik y R ! i t k gives
i
$ E D E á E E I á Ú !J% e $ E D E E %
e ik R na i k t na ik R i kt
(154)
Finding waves that can travel in this case is the same as repeating the calculation that led
to Eq. (128), but with U set to zero, since all bonds are broken, and with UN 0, since
phonons can propagate without any driving term. Examining Eq. (128), one sees that the
condition for surface phonons to propagate far behind the crack tip is F f, 0. Similarly,
far ahead of the crack tip no bonds are broken U should be set equal to U , and the condi-
tion for phonons is F
f>! 2 cos2 f 4c 0. According to Eq. (131), the roots and poles of
Q
f are therefore the phonon frequencies behind and ahead of the crack, and these are the
quantities appearing in Eq. (146). I do not know if Eq. (146) is more than approximately
correct for particle interactions more general than ideally brittle bonds.
42
5.0
4.0
Displacement U
3.0
2.0
1.0
0.0
-50 -25 0 25 50
Time t
Figure 17: Behavior of U t for Î 2, b 0 05, and N 9. Notice that at t 0, indicated
by the dashed line, u is decreasing, and that it had already reached height 1 earlier. This
state is not physical.
43
\ 3 2, and velocities will be scaled by this value. For N 9, at a velocity of c c
666 , c 1 158 , the bond between u 0 1 2 and u 1 1 2 reaches a distance of
2u 2r 2r
f some short time after the bond between u 0 1 2 and u 0 ! 1 2 snaps. The steady
state solutions strained with larger values of are inconsistent; only dynamical solutions
more complicated than steady states, involving the breaking of bonds off the crack path,
are possible. To investigate these states, one must return to Eq. (99) and numerically solve
the model directly. These simulations have been carried out [37, 48] and some results are
contained in Figure 18.
These theoretical calculations were done in response to experiments [49–51, 51] that
showed cracks The geometry of the instability is the same as the geometry of instabilities
found experimentally in Plexiglas, shown in Figure 19. However, a detailed connection
between the theory, which describes the onset of branching in a crystal, and the experiment,
which observes mesoscopic branching in an amorphous polymer, has not fully been worked
out.
K L 1 N 147
M
OQP c L 0 N 645
K L 1 N 165
M
OQP c L 0 N 630
K L 1 N 376
M
OQP c L 0 N 624
K L 1 N 835
M
OQP c L 0 N 775
Figure 18: Pictures of broken bonds left behind the crack tip at four different values of .
The top figure shows the simple pattern of bonds broken by a steady-state crack. At a value
of slightly above the critical one where horizontal bonds occasionally snap, the pattern is
periodic. All velocities are measured relative to the sound speed c \ 3 2. Notice that the
average velocity can decrease relative to the steady state, although the external strain has
increased. As the strain increases further, other periodic states can be found, and finally
states with complicated spatial structure. The simulations are carried out in a strip with
half-width N 9, of length 200 and b 0 01. The front and back ends of the strip have
short energy-absorbing regions to damp traveling waves. The simulation was performed by
adding unbroken material to the front and lopping it off the back as the crack advanced.
The diagram shows patterns of broken bonds left behind the crack tip. Just above the
44
Figure 19: Images of local crack branches in the x ! y plane in PMMA. The arrow, of length
250e m, indicates the direction of propagation. All figures are to scale with the path of the
main crack in white. (top) ÒÜà c (center) 1 18 c (bottom) 1 45 c . (from [52])
threshold at which horizontal bonds begin to break, one expects the distance between these
extra broken bonds to diverge. The reason is that breaking a horizontal bond takes energy
from the crack and slows it below the critical value. The crack then tries once more to reach
the steady state, and only in the last stages of the approach does another horizontal bond
snap, beginning the process again. This scenario for instability is similar to that known
as intermittency in the general framework of nonlinear dynamics [53] ; the system spends
most of its time trying to reach a fixed point which the motion of a control parameter has
caused to disappear. Heizler, Kessler, and Levine [43] have shown that when the snapping
bond is replaced by a rounded potential, the steady crack states are linearly unstable in a
more conventional way, the the instability is a Hopf bifurcation.
1. The force law between atoms is actually much more complicated than ideal snapping
bonds. Gao [54], has pointed out that the Rayleigh wave speed in the vicinity of
45
a crack tip may be significantly lower than the value of cR far away from the tip
because material is being stretched beyond the range of validity of linear elasticity.
2. The experiments are at room temperature, while the calculations are at zero.
3. The experiments are in amorphous materials, while the calculations are in crystals.
ð % 3ì ï
3a2
c2l aì (158a)
8m
ð % ì ï
3a2
ct2 3ì (158b)
8m
where m is the mass of each particle. In addition to forces between neighbors, it is possible
to add complicated dissipative functions depending upon particle velocities. In Ref. [38], a
term was added to the equations so as to reproduce the experimentally measured frequency
dependence of sound attenuation in Plexiglas. There is a slight technical restriction in
the calculations of Ref. [38]; right on the crack line, forces are required to be central.
46
0.8 Microcracking instability
Minimum velocity
0.6
steady state
65 Height=80
Height=160
ct
0.4
Velocity gap
0.2
Lattice trapping
0.0
1.0 1.2 1.4 1.6 1.8 2.0
Figure 20: Crack velocity versus loading parameter KI KIc for strips 80 and 160
atoms high, in limit of vanishing dissipation, computed for Mode I crack by methods of
following section. will be defined precisely in Eq. (108). The spring constants have
values ì ï 2 6 and ì ð 0 26, so that the non-central forces are relatively small. The
fact that the results have become independent of the height of the strip for such small
numbers of atoms in the vertical direction suggests that relatively small molecular dynamics
simulations can be used to obtain results appropriate for the macroscopic limit.
E. Gerde has found a way to overcome this technical limitation, and the results do not
change noticeably.
There is no universal curve describing Mode I fracture. Many details in the relation
between loading and crack velocity depend upon ratios of the sound speeds, and upon
the frequency dependence of dissipation. In the limit of central forces, ì ð 0, it turns
out to be difficult but not impossible to have cracks in steady state. This means that the
range of loads for which cracks can run in a stable fashion is small, and depends in detail
upon the amount of dissipation. Crack motion is greatly stabilized by having nonzero
ì ð . Figure 20 displays a representative result with some implications for the design of
molecular dynamics simulations. In the limit of vanishing dissipation, the relation between
crack speed and dimensionless loading KI KIc becomes nearly independent of the
number of vertical rows of atoms for strips around 40 rows high. The definition of is
that it describes the vertical displacement imposed upon the top and bottom of the strip, but
scaled so that 1 when these boundaries have been stretched just far enough apart so
that the energy stored per length to the right of the crack tip just equals the energy cost per
length of extending the crack. It is defined precisely by Eq. (108)
Analytical work has continued on these classes of models, and there are many results
I will not refer to much. Kessler and Levine have conducted the most careful studies of
lattices with nonlinear force laws and dissipation acting in concert [43, 44]. Their findings
emphasize the point that crack dynamics are terribly sensitive to features of the tip that do
appear in any continuum form of the problem. With Eric Gerde, I applied some of these
ideas to interface fracture and to friction [45].
47
5 Scaling ideas and molecular dynamics
Because fracture involves both large and small scales, because it is of practical importance,
and because it became fashionable for reasons that are hard to explain precisely, there is
an ongoing effort of many groups to study fracture through numerical methods that reach
down to the atoms. These studies fall into roughly two groups:
1. Calculations employing classical molecular dynamics. The largest studies employ a
billion atoms or more [55–59].
2. Calculations employing continuum mechanics, but with special spatial regions where
the analysis descends to molecular dynamics, and perhaps even, in small regions,
quantum mechanics [60–65]
My own prejudice is toward carrying out molecular dynamics in the spirit of fracture
mechanics. That is, instead of trying to simulate a macroscopic sample in its entirety
in the midst of one computer program, I prefer to use computer calculations to tabulate
quantities that linear elastic fracture mechanics needs in order to make predictions. If
one assumes that cracks move along a straight line, then what one then has to compute is
fracture energy as function of crack velocity. In what follows, I will show how this task can
be accomplished with numerical systems of relatively small size.
The essential idea is to abstract from the analytical solutions in lattices two scaling
laws, one for space and one for time. The time scaling law is easy. Simply look for steady
states. Once one finds numerically a steady state obeying the symmetry (109a), one knows
what it does forever, and extrapolating up to time scales of milliseconds is no problem.
The spatial scaling is slightly trickier. In order to relate samples of different size to one
another, let Gc be the Griffith energy density [12,66]; that is, twice the crack surface energy
density, a lower bound on the energy per unit area required for a perfectly efficient crack to
propagate along a certain plane. One can then define a dimensionless measure of loading
to be
G Gc
(159)
where G is the fracture energy density already introduced, i.e. the elastic strain energy
stored per unit area (in the fracture plane) ahead of the crack. Analytical solutions for the
ideal brittle solid show that the relationship between and crack velocity becomes in-
dependent of the height of the strip (number of planes stacked vertically) for surprisingly
small strips [67]; a strip 80 atoms high has for all practical purposes reached the infinite
limit, Fig 20. Guidance for conducting computationally expensive, moderately realistic,
molecular dynamics simulations can be had from this simple result: The very rapid conver-
gence of the main quantity of physical interest allows one to obtain physically meaningful
results from simulations that are considerably smaller than many being carried out these
days . This approach can also be viewed as an alternative to methods that join together
atoms and continua [60–65].
If computational resources were infinite, one would still need to choose properly the
sample geometry in order to compare with and inform experiment and theory. With limited
computational resources, choice of sample geometry and scale are of paramount impor-
tance: when performing molecular dynamics simulations of fracture, one should not ask
48
how large a system one can work with, but rather what is the smallest system that will
give results scalable to the macroscopic level. A good scaling argument can transform a
‘grand challenge’ problem into something that’s better focused and less of a challenge.
Size does matter: a smaller spatial scale allows one to follow the evolution of systems for
longer times; and with a strip geometry one can then analyze fully steady state behavior
and connect with theory and experiment.
F ma ( (160)
that is it consists of integrating Newton’s equations of motion for all particles in lock step
over a series of time steps, the size of the step being chosen small enough to give converged
dynamics. Time step integration is often done using the Verlet algorithm [68–70].
Lattice theory makes predictions that are hard to observe in experiment, and one of
the reasons for doing computer simulations is to relate the two. The simulations I will
describe here in detail were carried out in silicon. Silicon is extremely brittle, and high-
quality macroscopic single crystal wafers are cheap. It is therefore an excellent candidate
for laboratory fracture experiments. Silicon is also technologically of great importance,
and as a result is one of the most studied materials. One measure of this is that there are
over thirty effective (although rather ineffective) interatomic potentials for silicon in the
literature [71, 72]. Furthermore, transmission electron micrographs of cracks in silicon
wafers reveal atomically sharp crack tips [13,73]. Silicon therefore is an obvious candidate
for molecular dynamics investigations of dynamic fracture, and an appropriate setting for
testing atomic–level understanding. If there is any brittle element for which one ought to
have the knowledge base making it possible to calculate fracture properties in detail, and
the motivation to carry the work through, it would be silicon.
49
With effective potentials, 107 atoms followed for a few tens of nanoseconds is achiev-
able. This difference in computational scales becomes important when modeling processes
which require a minimum of 105 atoms to capture just some of the complex underlying
physics — processes involving fracture, dislocation loops, grain boundaries, or amorphous-
to-crystal transitions, for example.
Another motivation for constructing effective interatomic potentials is that they make
the complex physics of what are fundamentally quantum mechanical phenomena more
physically intuitive, so that one may interpret the results of atomistic simulations in terms
of simple principles of chemical bonding [72].
WYSkU
l!T
s urt
ommq pn VkU
ijT
S"U S!T T"U V"T WXU!U WYS!T
Z[Y\^]"_a`cbehdf g
Figure 21: Crack tip blunting in Stillinger-Weber silicon: two dislocations open up at the
tip, preventing it from advancing. The crack is pointing in the direction 1 1 0k in the plane
(110), and the system is loaded with a strain parameter 1 6.
i
50
the potential will not yield fracture along the experimentally preferred fracture planes 111
and 110 : At low or moderate strains a crack will not move at all. What happens is that two
dislocations open up at the crack tip, blunting it and preventing it from advancing, Fig 21.
One can play around with giving a transverse opening velocity to a select few atoms around
the tip. But to no avail. The crack simply will not crack. At very high strains, the crack tip
region melts.
Stillinger-Weber does give a type of fracture along the 100 plane which is quite rough
on the atomic scale, Fig 22. Abraham et al call this brittle fracture [58]. There is as yet
little consensus on a precise definition of brittleness. The experimental results of Lawn and
Hockey [13,73] for fracture along 111 in silicon show, however, that it is possible to have
atomically sharp fracture, i.e. where the newly created fracture surfaces are atomically
flat. The experimental evidence for fracture along 100 , on the other hand, is scant and
inconclusive [75], and that Stillinger-Weber yields fracture along this plane, albeit in a
rough manner, might even be yet another indication of the potential’s shortcomings.
It is possible to get cracks going, in an ideal brittle manner, along 111 and 110 with
the Stillinger-Weber potential by increasing the restoring forces between pairs of bonds,
i.e. by increasing the stability of the tetrahedra in the diamond lattice, and thus making
the crystal more brittle. This can be done by scaling a parameter, ± , the coupling constant
in the three-body term. Once ± has been changed in this way, the potential acquires a
new name, and is known as the Inadvertently Modified Stillinger-Weber potential (IMSW).
Originally, Stillinger and Weber set ±U 21 (dimensionless). However, by doubling this, as
in the pioneering inadvertent modification, one can obtain fast brittle fracture. With a fast
crack running, if one quasistatically decreases ± , the crack arrests well before one reaches
±z 21, Fig 23. Although ±z 42 gives fracture phenomenology in reasonable accord
a
a
Ia Ia Ia I
v w xXyaz|{~}k
Figure 22: A two-dimensional projection of rough cracking along the (100) plane in
Stillinger-Weber silicon. The system is loaded with a strain parameter 1 7, corre-
sponding to a fracture energy density of almost three times the Griffith energy density. The
average crack speed is 1.9 km/s.
51
F§ ¨Y©4§ ©
k
ÌÊ¹Ë ¥
ÈÀ·É
ÅÆ Ç ¢
ýÄ | ¡
À¯½¿Á
½¾
¢Y£ ¢X¤ ¥k ¥k¢ ¥ ¥k£ ¥k¤ ¢
¦
§«ª^©4§ ¢jY ©
ÌÊ¹Ë ¥
ÈÀ·É
Å Æ ½Ç ¢
ÃÄ || ¶ ¬·¬ ¯³¹®±¸?° º ²»³ ¬·¯®±°¼² ³
À¯½¿Á ´ µ ¬¯®±° ² ³
½¾
¢Y¢ ¢ ¢X£ ¢X¤ ¥k ¥"¢ ¥ ¥Y£ ¥k¤ ¢
¦
Figure 23: Crack velocity profile (a) along (110) 11 0k , and (b) along (211) 11 1 k in modi-
i i
fied Stillinger-Weber silicon: ± scales the strength of angular forces between pairs of bonds
that stabilize the diamond lattice structure. Stillinger and Weber’s value for ± is 21. How-
ever, leaving all other parameters unchanged, cracks will not propagate unless one uses a
larger value of ± . specifies the loading.
with experiment, it has the adverse effect of raising the melt temperature above 3500 K,
whereas experimentally the melt happens at 1685 K. The Young moduli also get shifted.
Moduli results from tensile tests on small samples, for three lattice directions, are given in
Table 2 for both the original and modified Stillinger-Weber (SW) potentials, along with the
corresponding experimental values.
The story of searches for potentials after learning about the Stillinger-Weber potential
and its inadvertent modifications is long and alternately depressing and amusing. Dominic
Holland and I devoted a fair amount of attention to a potential of Kaxiras and Bazant
[72,76–78], and then one of Chelikowsky and Phillips [79]. We learned that these potentials
and many others have a defect that can easily be understood in a qualitative way.
Imagine taking a uniform crystal and simply expanding the lattice constant. This is a
52
computation that can be done easily by density functional theory, and one can compare
the results with the same calculation repeated with a classical potential. Then repeat the
computation putting the crystal in uniform shear. Most classical potentials have too large
a peak when the crystal is expanded, and resist shearing too little. The height of the peak
resisting expansion can be as much as 5 times larger than the corresponding peak predicted
by density functional theory. Getting these features of atomic force laws wrong is terrible
if one wants to predict fracture properties at all accurately. If shearing is easier than pulling
apart in tension, a crystal will emit dislocations by shearing at the tip rather than breaking
in tension.
In searching for a potential that would correctly reproduce the qualitative features ob-
tained from density functional theory, I eventually settled upon the Modified Embedded
Atom Method of Baskes [80, 81]. This potential is based upon a number of physical ideas
that I find quite persuasve, particularly the idea of screening. The density functional results
for uniform expansion are built into the potential from the beginning, Without any tuning
or adjustment, it does an acceptable job of describing the resistance to shear, and in com-
puter simulations it does permit the motion of brittle cracks. Still, there are some arbitrary
features of the potential that are a bit unnerving, and from a quantitative point of view it
does not turn out in the end to be any better than the IMSW with which our studies began.
E á E t%
zR na E
na I4, zR t
(161)
53
6
5 Bazant-Kaxiras
4
Force (eV/Å)
3 Stillinger-Weber
2 Density functional
(A)
1.0 1.2
Î
a a0
1.4 1.6
10
Density Functional Theory
Modified Stillinger Weber
8 Original Stillinger Weber
Modified Embedded Atom Method
Energy per atom (eV)
0
1 1.5 2 2.5 3
(B) Extension Ratio
Figure 24: (A) Comparison of density functional theory, EDIP, and the Stillinger-Weber po-
tential for the restoring force during uniform expansion. MEAM is identical to the density
functional result by construction. (B) Same as (A), but now repeated for uniform shear.
For a range of loads , Eq. 161 applies for any pair of atoms, whatever their separation
along the crack surface. In order to obtain the perfect periodicity shown in Fig. 25, the
crack was allowed to run first for 60,000 time steps so as to come into equilibrium with the
waves it sends towards top and bottom boundaries.
The longer and/or higher the system the longer it will take to reach a steady state. The
240 ps required to reach the steady state depicted in Fig 25 is about an order of magnitude
longer than the duration of most large-scale molecular dynamics simulations of fracture.
Using a reduced spatial system size, as validated by the scaling argument, is what makes
this possible. This becomes crucially important when proceeding through a sequence of
steady states, which requires quasistatic changes in the loading, or G. To obtain a full
set of results, like Fig 26, a crack actually will travel tens of microns, or for times on the
order of a tenth of a microsecond. To achieve this, one needs not only efficient code and
a high performance computer, but also a physically motivated smallest computational cell
— a minimum thin strip on a conveyor belt.
54
Steady State after 0.24 ns at ∆=1.6
vertical profile of two atoms 160 Angstroms apart
5
0
1 2 3 4 5
time (ps)
Figure 25: Crack that has been traveling for over 0.24 ns. That the two overlapped
curves are almost completely indistinguishable shows the crack has reached steady state,
according to Eq. 161, and is emitting phonons in accord with Eq. 153.
üký"þæÿ
#
"
"
!
Figure 27: Relation between crack speed and loading for cracking along 111 01 1 k at
i
0 K. Dotted lines indicate forbidden velocities. The lower figure shows crackons for 1 18 Ü
Ü 1 44: the crack is able to expose many%$different states lying along many hysteresis
loops. Ideal steady states are unstable above 2 2.
56
for ê 2 2 it goes unstable in a similar manner to cracks along 110 . However, for
1 175 Ü
that have notÜ been1 44 the dynamics of the crack exhibits a number of interesting features
seen previously, and for which there is not yet a complete theoretical
description. There is a variety of different dynamical states available for each value of ,
where the crack travels at different speeds. Each of these states corresponds to a plateau
in ; can change by as much as one fifth of the amount needed to go from arrest
to instability and the crack velocity does not alter within numerical resolution. When the
crack finally decides to accelerate out of the plateau, it may jump by over 1 km/s and reach
an upper plateau to within a few m/s. On cyclical loading the same plateaus are always
reached. All of these transitions are hysteretic, as depicted in Fig. 27. The different states
emit noticeably different phonons; on a given plateau, the phonon frequencies appear fixed
and their amplitude changes, while between plateaus the frequencies change in accord with
Eq. 153. This is crackon behavior. All these phenomena are easily disguised if strain rates
are too high. Resolving all the fine structure visible in Fig. 27 required Ü 8 e s 1 , or *
*ä
hz c Ü 10 5 .
ä
4
3 unstable
crack speed (km/s)
0
1.2 1.4 1.6 & 1.8 2.0 2.2
57
/
) 476
) 82496
4
L
IKJ ) ).404:6
>H
G 4
F
) )8+4:6
C D ;E
A,@B 4
>?
; ) /2404:6
=
;<
4
) /+8+4:6
) 3+404:6
4
)+*,).- )+* /+/ )0* /21 )+* 3+4 )+* 325
(
Figure 29: Velocity gap vanishes near 200 K. Dotted lines indicate forbidden velocities.
All low-lying velocity states exist at room temperature. Cracking is along (111) 01 1 k .
i
5.8 Comparisons with experiment
I turn to a number of questions that allow a comparison of theory and experiment. First,
is there a velocity gap in brittle amorphous materials at room temperature? The answer
seems pretty clearly to be “no;” data to this effect appear in Figure 31. There is no de-
tailed dynamical theory for the motion of cracks in very brittle amorphous materials, so the
experiments are the best guide to what happens.
However, the point of performing numerical work in crystalline silicon was the possi-
bility of carrying out experiments to compare directly. The first were done by Hauch et
al. [82], with additional experiments by Cramer, Wanner, and Gumbsch [83], and more re-
cently by Deegan, who has been working with me to obtain data at cryogenic temperatures.
Room-temperature runs were performed along 111 1 1 2k in order to allow direct com-
i
parison with experiment [82] (the zero Kelvin runs for this lattice direction are in Fig 28).
½ ½
The results are in Fig. 32, and were obtained for a thin strip of size 532 15 154 Å ,
3
periodic along the thin axis. As before, new material was added ahead of the crack tip and
old material lopped off at the tail every time the crack advanced to within 200 Å of the
forward end of the strip. In this fashion, the crack traveled 7 e m during the course of the
simulations as G was varied between 5 and 14 J/m2 . The room-temperature experiments
for fracture in silicon wafers along (111) 1 1 2k are described in [82]. These experiments
i
are difficult to perform because of the high Young’s modulus and brittleness of silicon.
58
S
si~ Q
} P.N m
|
y z p{
wixv P.N M
P
sut si~
p }
pqr |
M2N m
M
M2N M
b ^jW b ^YkWhcU+l P.N Q\S P.N Q
O P.N oQ
_
P
n
M+N O PN Q P.N R QN M Q2N S
_
Figure 30: Crack velocity profile with respect to quasistatic loading along 111 01 1 k in
silicon at 300 K. All low-lying velocity states exist.
i
300
PMMA
200
100
0
0 100 200 300
300 Homalite-100
(m/sec)
200
100
Velocity
0
0 100 200 300 400 500
75
Glass
50
25
0
Figure 31: Velocity records for experiments in the double cantilever beam configuration. In
PMMA and Homalite-100 crack arrest was achieved. In both materials the crack appears
to decelerate smoothly but with increasing deceleration until it arrests. In glass crack arrest
was not achieved in this configuration. However there is no sign of an initial velocity jump.
This can be attributed to the sharp seed cracks that can be generated in glass.
59
6
4
Length [cm]
4
3
2
(km/s)
0
0 2 4 6 8 10 12 14 16 18 2
4
Velocity [km/s]
3 Experiment
1
2
Molecular Dynamics
1
0
0 0 5 10 15 20
0 2 4 6 8 10 12 14 16 18
Energy flow to crack tip G [J/m2]
(A) Time[µs]
(B)
4
Velocity [km/s]
2
Experiment (111, Hauch)
Experiment (110, Cramer)
Experiment (111, Deegan, ramp)
1
0
0 20 40 60 80
2
G [J/m ]
(C)
Figure 32: Experimental and numerical determination of the crack velocity as a function
of fracture energy G along 111 1 1 2k at room temperature. The fuzz indicates the spread
i
in simulation velocity data. Experimental data from Hauch et al. [82], Cramer et al. [83]
and R. Deegan.
60
The highest experimental and numerical crack velocities shown in Fig. 32 are reason-
ably close, but the minimum fracture energies at which a crack propagates differ: 2.3 J m 2
in the experiments and 5.2 J m2 in the simulations. Since the scale of crack velocities in
a material is bounded by sound speeds [12], which are given correctly by the Stillinger-
Weber potential, it is not surprising that the experimental and computational crack velocity
scales agree. Furthermore, the potential gives the correct cohesive energy of silicon (but
an inaccurate cohesive energy curve — see Sec 5.3), leading to the agreement in numerical
and experimental energy scales. However, the nonlinear parts of the potential involved in
stretching and rupturing bonds play an important role in determining the actual fracture
energies and crack velocities, in particular where the crack arrests and what its highest ve-
locity will be. The quantitative disagreements shown here point to a shortcoming of the
nonlinear parts of the potential, which have not received much attention. The modified
Stillinger-Weber potential cannot be expected to reproduce correctly experimental data.
But it highlights the control over brittleness in the three-body term, and the inadequacy of
the potential tails in the two-body term. These are complicated matters, and yet perhaps
only hinting at greater difficulties on the road to a better potential.
The lowest fracture energy density at which a crack propagated in the experiments was
close to the Griffith energy density for a (111) plane, 2.2 J/m2 [84]. Since a crack cannot
travel with less energy, there must be a narrow range of fracture$ energy over which the crack
velocity rises rapidly from zero to the lowest value measured, 2 km/s. This phenomenon
is also seen in glass and polymers [38]. Because of the extreme precision required at the
boundaries, experiments with silicon are not yet capable of settling the matter of whether
this sharp velocity rise signals a velocity gap. However, as shown in Fig 29, in numerical
silicon, the velocity gap is temperature dependent, vanishing above 200 K. At 300 K there
do exist ‘steady states’ at all velocities between 0 and 3 km/s.
The silicon experiments covered a range of fracture energy densities, 2 ! 16 J m 2 , in
which the cracks produced very smooth surfaces. Thus, cracks in silicon can dissipate large
amounts of energy, more than seven times the amount needed to create a clean cleavage
through the whole crystal, without leaving behind any large scale damage on the fracture
surfaces. Investigation by atomic force microscopy shows that for low fracture energies
the fracture surfaces are flat on the nanometer scale, while at higher energies the surfaces
have pronounced features. These features, however, are smooth on the micron scale, and
account for height variations on the order of 30 nm over an area of 16 e m 2 . The roughness
gives an area increase of only 0.1% above that of a flat cleaved surface. This extra surface
cannot account for the sevenfold increase in dissipated energy. The simulations, however,
indicate that most of the energy can be carried off in lattice vibrations.
61
Theory and experiment in Silicon
2
MEAM, 0 kelvin
MEAM, 77 kelvin
1 Hauch et al, (111)
IMSW, 300 kelvin
Cramer et al, (110)
0 Deegan, (111)
0 2 4 6 8 10 12
2
Fracture energy [J/m ]
Figure 33: Molecular dynamics runs at 300 K with Stillinger-Weber potentials and 0K and
77 K runs compared with experiments of Hauch et al, Cramer, Wanner, and Gumbsch [83],
and Deegan.
Figure 34: Atomic force microscopy pictures of (111) fracture surfaces in silicon at three
different energies: (a) G 4 3 J/m2 , (b) G 7 2 J/m2 , and (c) G 14 4 J/m2 . The crack
has traveled from left to right in a 1 1 2k direction. The diagonal surface markings in (b)
i
and (c) are approximately in a 11 0k direction, but it is unclear what causes them. Pictures
i
by Rachel Mahaffy; experiments and figure composition by Jens Hauch.
I will close by commenting on how well they are yet answered. For ideal brittle crystals,
such as silicon, the energy needed to propagate can now be calculated. The scaling prob-
lems needed to find the effect of microscopic features on macroscopic behavior have been
solved by learning from analytically solvable models. However, quantitative comparison of
theory and experiment is not better than around a factor of two. Furthermore, in amorphous
materials such as glass there is no theory, and in more complex materials with intermediate
microstructures it is not clear what predictive value atomic–scale considerations bring.
The stability of the ideal brittle crystals against microscopic branching can be deter-
mined. However, the instabilities observed in real materials are on much larger than atomic
scale, and they have not been followed far or explained.
62
Finally concerning crack paths, there are phenomenological rules that appear to have
much experimental support for cracks that travel slowly in isotropic materials. For cracks
that move quickly, it is quite uncertain what the rules ought to be, and neither theory nor
experiment have gone very far.
The interest and difficulty of these problems should be sufficient to ensure that fracture
continues to interest physicists for some time to come.
References
[1] B. L. Averbach, Fracture, an Advanced Treatise, vol. I, ch. Some physical aspects of
fracture, pp. 441–471. New York: Academic Press, 1968.
[4] N. Mott, “Brittle fracture in mild steel plates,” Engineering, vol. 165, pp. 16–18, 1947.
[5] E. Dulaney and W. Brace, “Velocity behavior of a growing crack,” Journal of Applied
Physics, vol. 31, pp. 2233–2266, 1960.
[7] E. H. Yoffe, “The moving Griffith crack,” Philosophical Magazine, vol. 42, pp. 739–
750, 1951.
[9] L. I. Slepyan, Models and Phenomena in Fracture Mechanics. Berlin: Springer, 2002.
[10] K. B. Broberg, Cracks and Fracture. San Diego: Academic Press, 1999.
[11] M. F. Kanninen and C. Popelar, Advanced Fracture Mechanics. New York: Oxford,
1985.
[13] B. Lawn, Fracture in Brittle Solids. Cambridge: Cambridge University Press, sec-
ond ed., 1993.
[14] D. Broek, The Practical Use of Fracture Mechanics. Dordrecht: Kluwer Academic
Press, 3rd ed., 1994.
63
[15] G. R. Irwin, “Fracture mechanics,” in Structural Mechanics (J. N. Goodier and N. J.
Hoff, eds.), pp. 557–591, Elmsford, NY: Pergamon Press, 1960.
[16] E. Orowan, “Energy criteria of fracture,” Weld. Res. Supp., vol. 34, p. 157, 1955.
[17] B. Cotterell and J. R. Rice, “Slightly curved or kinked cracks,” International Journal
of Fracture, vol. 14, p. 155, 1980.
[18] G. R. Irwin, “Fracture,” in Handbuch der Physik, vol. 6, p. 551, Berlin: Springer-
Verlag, 1958.
[19] G. R. Irwin, “Analysis of stresses and strains near the end of a crack traversing a
plate,” Journal of Applied Mechanics, vol. 24, pp. 361–364, 1957.
[21] H. Gao, Y. Huang, and F. F. Abraham, “Continuum and atomistic studies of intersonic
crack propagation,” JMPS, vol. 49, pp. 2113–32, 2001.
[22] A. J. Rosakis, “Intersonic shear cracks and fault ruptures,” Advances in Physics,
vol. 51, pp. 1189–257, 2002.
[23] R. V. Gol’dstein and R. Salganik, “Brittle fracture of solids with arbitrary cracks,”
International Journal of Fracture, vol. 10, pp. 507–523, 1974.
[25] A. Yuse and M. Sano, “Transition between crack patterns in quenched glass plates,”
Nature, vol. 362, pp. 329–31, 1993.
[26] O. Ronsin, F. Heslot, and B. Perrin, “Experimental study of quasistatic brittle crack
propagation,” Physical Review Letters, vol. 75, pp. 2252–5, 1995.
[27] M. Marder, “Instability of a crack in a heated strip,” Physical Review E, vol. R51,
pp. 49–52, 1994.
[29] M. Adda-Bedia and Y. Pomeau, “Crack instabilities in a heated glass strip,” Physical
Review E, vol. 52, pp. 4105–4113, 1995.
64
[31] H. Larralde and R. C. Ball, “The shape of slowly growing cracks,” Europhysics Let-
ters, vol. 30, pp. 87–92, 1995.
[32] J. R. Rice, Fracture, vol. 2, ch. Chapter 3, pp. 191–311. Academic Press, 1968.
[34] A. Karma, D. Kessler, and H. Levine, “Phase-field model of Mode III dynamic frac-
ture,” Physical Review Letters, vol. 87, p. 045501, 2001.
[36] L. Slepyan, “Dynamics of a crack in a lattice,” Soviet Physics Doklady, vol. 26,
pp. 538–540, 1981.
[37] M. Marder and X. Liu, “Instability in lattice fracture,” Physical Review Letters,
vol. 71, pp. 2417–2420, 1993.
[38] M. Marder and S. Gross, “Origin of crack tip instabilities,” Journal of the Mechanics
and Physics of Solids, vol. 43, pp. 1–48, 1995.
[39] R. Thomson, C. Hsieh, and V. Rana, “Lattice trapping of fracture cracks,” Journal of
Applied Physics, vol. 42, no. 8, pp. 3154–3160, 1971.
[40] R. Thomson, “The physics of fracture,” Solid State Physics, vol. 39, pp. 1–129, 1986.
[41] D. Holland and M. Marder, “Ideal brittle fracture of silicon studied with molecular
dynamics,” Physical Review Letters, vol. 80, pp. 746–9, 1998.
[44] D. A. Kessler and H. Levine, “Nonlinear lattice model of viscoelastic mode III frac-
ture,” PRE, vol. 63, pp. 016118/1–9, 2001.
[45] E. Gerde and M. Marder, “Friction and fracture,” Nature, vol. 413, pp. 285–288, 2001.
[46] B. Noble, Methods Based on the Wiener-Hopf Technique for the Solution of Partial
Differential Equations. New York: Pergamon, 1958.
[48] X. Liu, Dynamics of fracture propagation. PhD thesis, University of Texas at Austin,
1993.
65
[49] K. Ravi-Chandar and W. G. Knauss, “Dynamic crack-tip stress under stress wave
loading-a comparison of theory and experiment,” International Journal of Fracture,
vol. 20, pp. 209–222, 1982.
[50] K. Ravi-Chandar and W. Knauss, “An experimental investigation into dynamic frac-
ture: I. Crack initiation and arrest,” International Journal of Fracture, vol. 25,
pp. 247–262, 1984.
[52] E. Sharon, S. P. Gross, and J. Fineberg, “Local crack branching as a mechanism for
instability in dynamic fracture,” Physical Review Letters, vol. 74, pp. 5146–5154,
1995.
[53] P. Manneville, Dissipative Structures and Weak Turbulence. Boston: Academic Press,
1990.
[54] H. Gao, “Surface roughening and branching instabilities in dynamic fracture,” Journal
of the Mechanics and Physics of Solids, vol. 41, pp. 457–486, 1993.
[57] F. F. Abraham, “Portrait of a crack: rapid fracture mechanics using parallel molecular
dynamics,” IEEE Computational Science and Engineering, vol. 4, pp. 66–77, 1997.
66
[62] H. Rafii-Tabar, L. Hua, and M. Cross, “A multi-scale atomistic-continuum modelling
of crack propagation in a two-dimensional macroscopic plate,” vol. 10, pp. 2375–,
1998.
[66] A. Griffith, “The phenomena of rupture and flow in solids,” Mechanical Engineering,
vol. A221, pp. 163–198, 1920.
[69] D. Frenkel and B. Smit, Understanding Molecular Simulation. New York: Academic
Press, 1996.
[73] B. Lawn, B. Hockey, and S. Wiederhorn J. Mater. Sci., vol. 15, p. 1207, 1980.
[77] J. Justo, M. Z. Bazant, E. Kaxiras, V. Bulatov, and S. Yip Physical Review B, vol. 58,
p. 2539, 1998.
67
[78] E. K. M. Z. Bazant and J. F. Justo, “The environment-dependent interatomic potential
applied to silicon disordered structures and phase transitions,” Mat. Res. Soc. Proc.,
vol. 491, p. 339, 1997.
[79] J. Chelikowsky and J. Phillips Physical Review B, vol. 41, p. 5735, 1990.
[80] R. Ravelo and M. I. Baskes, “Free energy calculations of Cu-Sn interfaces,” in Ther-
modynamics and Kinetics of Phase Transformations. Symposium (J. S. Im, B. Park,
A. L. Greer, and G. B. Stephenson, eds.), (Pittsburgh), pp. 287–93, Materials Research
Society, 1996.
[81] M. I. Baskes, “Modified embedded-atom potentials for cubic materials and impuri-
ties,” PRB, vol. 46, pp. 2727–2742, 1992.
[83] T. Cramer, A. Wanner, and P. Gumbsch, “Energy dissipation and path instabilities in
dynamic fracture of silicon single crystals,” PRL, vol. 85, pp. 788–91, 2000.
[84] J. C. H. Spence, Y. M. Huang, and O. Sankey, “Lattice trapping and surface recon-
struction for silicon cleavage on (111). ab initio quantum molecular dynamics calcu-
lations,” Acta Metallurgica, vol. 41, pp. 2815–2824, 1993.
68