You are on page 1of 26

Plane-strain propagation of a uid-driven fracture during

injection and shut-in: Asymptotics of large toughness


Dmitry I. Garagash
*
Department of Civil and Environmental Engineering, Clarkson University, 230 Rowley Laboratories, 8 Clarkson Ave.,
Potsdam, NY 13699-5710, USA
Received 1 September 2004; received in revised form 14 June 2005; accepted 6 July 2005
Available online 2 November 2005
Abstract
This paper considers the problem of plane-strain uid-driven fracture propagating in an impermeable elastic medium
under condition of large toughness or, equivalently, of low fracturing uid viscosity. We construct an explicit solution for a
fracture propagating in the toughness-dominated regime when the energy dissipated in the viscous uid ow inside the
fracture is negligibly small compared to the energy expended in fracturing the solid medium. The next order corrections
in viscosity to this limiting solution are then derived, allowing the range of problem parameters corresponding to the
toughness-dominated regime to be established. The rst-order small viscosity (large toughness) solution is shown to pro-
vide an excellent approximation of the solution for the crack length in the wide range of the viscosity parameter. Further-
more, this solution, when combined with the rst-order small-toughness solution of Garagash and Detournay [Journal of
Applied Mechanics, 2005], provides a simple analytical approximation of the crack length solution in practically the entire
range of viscosity (toughness). It is also shown that the established method of asymptotic expansion in small parameter is
equally applicable to study other small eects (e.g., uid inertia) on the otherwise toughness-dominated solution. A solu-
tion for the fracture evolution during shut-in (i.e., after uid injection rate is suddenly stopped) is also obtained. This solu-
tion, which corresponds to a slowing fracture evolving towards the toughness-dominated steady state, draws attention to
the possibility of substantial fracture growth after uid injection is ceased especially under conditions when the fracture
propagation during injection phase is dominated by viscous dissipation.
2005 Elsevier Ltd. All rights reserved.
Keywords: Hydraulic fracture; Toughness; Inertia; Asymptotic solutions; Scaling
1. Introduction
The problem of a uid-driven fracture in rock arises in various applications ranging from hydraulic frac-
turing treatment used in the oil industry to stimulate oil production from underground reservoirs [1] to the
formation of intrusive dykes in the earth crust and magma transport in the lithosphere [2]. Other applications
0013-7944/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engfracmech.2005.07.012
*
Tel.: +1 315 268 6501; fax: +1 315 268 7985.
E-mail address: garagash@clarkson.edu
Engineering Fracture Mechanics 73 (2006) 456481
www.elsevier.com/locate/engfracmech
include stimulation and heat extraction from geothermal reservoirs, induced caving in the mining industry, soil
grouting, etc. The problem has attracted numerous contributions since the 1950s, especially in relation to
hydraulic fracturing treatments in the petroleum industry. These investigations have addressed various aspects
of the problem both analytically and numerically, see Ref. [3] for some further references. It is, however, only
recently [411] that there has been a rigorous eort to study the parametric dependence of a uid-driven frac-
ture and the corresponding dierent limiting regimes. This is partly due to the complex mathematical nature of
the fracture models, even for fractures of simple geometry (planar or disk-shaped) propagating in a homoge-
neous rock with a uniform in situ stress. Indeed, a consistent model of a hydraulic fracture has to capture the
non-local deformation of the crack with respect to the applied uid loading, the non-linear ow of viscous
uid in the fracture, and, in general, the presence of an unknown lag between the uid and the fracture fronts
and also the history-dependent leak-o of the fracturing uid into the permeable rock.
This paper investigates the limiting regimes of a planar hydraulic fracture propagating in an impermeable
linear elastic rock under conditions of negligible uid lag. These limiting regimes correspond to the dominance
of one of the two energy dissipation mechanisms related to the rock resistance to fracture (rock toughness) and
the uid resistance to ow (uid viscosity). In general, the propagation of a hydraulic fracture is controlled by
a dimensionless toughness /, which is related to the ratio of the energy expended in fracturing the solid to the
energy dissipated in the viscous uid [12].
In the viscosity-dominated regime (small /), Garagash and Detournay [13,11] have shown that (i) depar-
ture of the overall solution from the zero-toughness solution [8,14] is measured by S(/) = B
1
/
b
with b 3.18
and B
1
0.1; and (ii) that for S(/) 1 the toughness eects are localized to a boundary layer near the frac-
ture front of thickness /
6
(where is the fracture half-length) while the outer solution is given by the zero-
toughness solution. For moderate values of the dimensionless toughness, S(/) ~ 1, the solution departs from
the zero-toughness solution and its dependence on / cannot be neglected. The numerical method originally
proposed by Spence and Sharp [6] and later in rened form by Adachi [15] is appropriate to nd the solution
in this intermediate regime, where the eects of uid viscosity and rock toughness are of the same order. How-
ever, as / increases, the fracturing uid viscosity has to become eventually irrelevant, /(/) 1, where
/(/) is a dimensionless viscosity parameter with /() = 0, and the limiting solution corresponding to
the inviscid uid limit has to hold. The zero-viscosity solution of fracture propagation in the latter, tough-
ness-dominated regime, small-viscosity corrections to this solution, and parametric range for the regime are
the focus of this paper.
Hydraulic fractures in petroleum reservoir stimulation often considered to be viscosity-dominated
(S(/) 1) due to the high viscosity of industrial fracturing uids, long extent of these fractures, and rela-
tively high pumping rates. On the other hand, hydraulic fractures in the laboratory scale experiments are often
dominated by material toughness with barely any inuence from viscosity, even when the fracture is driven by
a very viscous uid [16]. Some other applications of hydraulic fracturing involving less viscous fracturing uids
(e.g., water in the heat extraction from geothermal reservoir) would also warrant the relevance of the tough-
ness-dominated regime. The criterion for the smallness of viscosity or toughness eects based on analytical (or
semi-analytical) asymptotic expansions of the solution and dimensionless analysis allow to determine (i)
whether or not the above or other regime assertions hold, and (ii) why. In other words, what is the dimension-
less parameter(s) characterizing the eect of toughness and viscosity and what is the range of the parameter
corresponding to a given limiting regime.
Propagation of a plane-strain fracture under condition of negligible fracturing uid viscosity has been pre-
viously considered by Huang et al. [17]. They have developed a self-similar solution under assumption of the
dominance of the uid inertia forces (as compared to the viscous drag) and for a particular injection law char-
acterized by the injection rate power-law t
1/5
in time. Other investigators [6,4] have assumed that the uid iner-
tia eects on either fracture propagation or the uid ow in the crack are negligible (even under conditions
when uid viscosity vanishes) and the uid ow can be modeled by the lubrication theory [18]. In this paper
we present the scaling for the fracture propagation driven by inertial, unidirectional ow of viscous uid, and
evaluate the eect of inertia. Based on this analysis we can maintain the physical relevance of the inertia-less
ow [6] at least in the context of conventional hydraulic fracturing treatments in the oil and gas industry, and
proceed to consider the plane-strain uid-driven fracture propagation characterized by small uid viscosity
(large material toughness) parameter. The perturbation method suggested here for evaluating small viscosity
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 457
eects on the otherwise toughness-dominated solution of a plane-strain hydraulic fracture is equally applicable
to either other fracture geometries and/or to evaluate other small eects (such as uid inertia) on the solution.
The paper is organized as follows. Firstly, we summarize the model describing a plane-strain uid-driven
fracture and discuss the scaling of the sought solution for a general uid injection law, the relevance of the
inertia-less assumption, and the asymptotic zero-viscosity solution of a fracture propagating in the tough-
ness-dominated regime. The next-order small-viscosity (large-toughness) solution is then obtained for two
injection laws, namely, a constant injection rate and the shut-in (when injection rate is abruptly reduced to
zero and the injected volume remaining constant thereafter), via an asymptotic series expansion in small
dimensionless viscosity, /(/) = /
4
. (The utility of the series expansion method to evaluate small eects
on the zero-viscosity solution other than that of the viscosity is demonstrated in Appendix C for the case
of small uid inertia.) We conclude with the discussion of the parametric dependence of a uid-driven fracture
in view of the two limiting propagation regimes.
2. Problem formulation
Consider propagation of a nite two-dimensional fracture of half length (t) in an impermeable linear elas-
tic medium characterized by Youngs modulus E, Poissons ratio m, and toughness K
Ic
. An incompressible uid
of viscosity l is injected at the center of the fracture at a rate Q(t), see Fig. 1, and the volume of injected uid is
given at any time t, V (t) =
_
t
0
Qdt. The crack is loaded by the internal uid pressure p
f
(x, t) and by the far-eld
conning stress r
0
. The fracture is assumed to be in mobile equilibrium at all times and its propagation is
quasi-static. We will search for the solution of the uid-driven fracture problem: the net pressure in the frac-
ture p = p
f
(x, t) r
0
, the crack opening w(x, t), and the fracture half-length (t), where x is the position along
the crack and t is the time.
Main assumptions underlining the considered model of plane-strain fracture propagation are summarized
as follows:
The fracture is fully uid-lled at all times, i.e., there is no lag between the fracture and uid front. This
assumption is particularly relevant at or near the toughness-dominated propagation regime, when the frac-
ture tip is blunt and the lag is an exponentially small function of toughness [9].
Fracture propagation is described in the framework of linear elastic fracture mechanics (LEFM), charac-
terized by the stress singularity at the fracture tip and the propagation criterion in mobile equilibrium
requiring the stress intensity factor to be equal to the material toughness K
Ic
[19]. The LEFM assumption
requires negligible inelastic deformations away form the fracture front and the smallness of the process zone
near the fracture front (wherein the inelasticity is assumed to be localized). The process zone at the fracture
front bears some similarity with the uid lag behind the fracture front, as both regularize otherwise singular
elds of stress and uid pressure, respectively. The eect of the process zone on fracture propagation can
generally be neglected (or represented via the eective fracture toughness [20]) when its dimension is small
compared to the fracture extent and/or under the condition of large conning stress.
Fig. 1. Sketch of a plane-strain uid-driven fracture.
458 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
The ow of incompressible uid in the fracture is unidirectional and laminar. These assumptions are jus-
tiable given the smallness of the crack opening compared to the crack length and small opening gradient
[6], and low Reynolds number of the ow [4].
No uid exchange between the fracture and the surrounding solid is taking place.
Under above assumptions, the governing equations are formulated over the half of the crack, 0 6 x 6 , on
the account of the problem symmetry. The formulation makes use of the following eective material
parameters:
E
/
=
E
1 m
2
; l
/
= 12l; K
/
= 4
2
p
_ _
1=2
K
Ic
; (1)
with the meaning of plane-strain elastic modulus, eective viscosity and toughness, respectively.
The uid ow in the fracture is governed by continuity of mass and momentum. An integral form of the
local continuity equation and the global uid balance, stating that the injected uid volume V(t) is equal to
the fracture volume, are given by
o
ot
_

x
wdx = wv;
_

0
wdx =
1
2
V (t); (2)
where v(x, t) is the uid mean velocity (in the cross-section normal to the ow) and wv is the uid ow rate per
unit (out-of-plane) width of the crack. In writing Eq. (2)
a
, the uid velocity at the fracture tip, v(x = ) has
been taken equal to the tip velocity d/dt, following the zero uid lag assumption. The momentum balance
equation for unidirectional laminar ow at small or very large values of Reynolds number is [21,4]
ov
ot

1
2
ov
2
ox
=
1
q
op
ox

l
/
v
w
2
_ _
; (3)
where l
/
v/w
2
is the Poiseuille expression for the viscous shear stress in the laminar channel ow. The lubrica-
tion ow approximation [18]
v =
w
2
l
/
op
ox
(4)
results from Eq. (3) when inertia terms (the left hand side of Eq. (3)) are neglected. Nilson [4,5] and Spence and
Sharp [6] argue for the lubrication approximation [Eq. (4)], whereas Huang et al. [17] consider inertial formu-
lation [Eq. (3)] and neglect the viscous shear stress l
/
v/w
2
. In the next section we analyze the scaling of the
problem and formulate the parametric regimes where one or the other assumption can be used.
The solid deformation and the fracture propagation criterion are prescribed by equations of linear elastic
fracture mechanics. The crack opening w is related to the net pressure p by an integral equation [22],
w(x; t) =
4
pE
/
_

0
G
x

;
x
/

_ _
p(x
/
; t)dx
/
; G(n; n
/
) = ln

1 n
2
_

1 n
/2
_

1 n
2
_

1 n
/2
_

. (5)
The criterion of continuous quasi-static propagation of a fracture in mobile equilibrium, K
I
= K
Ic
, is expressed
as the tip asymptote of the crack opening [19]
w =
K
/
E
/
( x)
1=2
x . (6)
Eqs. (2), (3) or (4), and (5), (6) fully dene the fracture length (t), uid velocity v(x, t), the opening w(x, t), and
the net pressure p(x, t) as functions of the parameter set [Eq. (1)], the uid density q, and the injection law V(t).
3. Scaling
A particular insight into the dierent regimes of hydraulic fracture propagation and nature of correspond-
ing solutions (e.g., self-similarity) can be gained from scaling considerations [4,6,23,9,24]. Following
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 459
Detournay [24], let us introduce the scaled coordinate n = x/(t) [0, 1] and the following dimensionless
quantities: the opening X, the net pressure P, the crack half-length c, and the uid velocity # as
w(x; t) = e(t)L(t)X(n; t); p(x; t) = e(t)E
/
P(n; t); (7)
(t) = L(t)c(t); v(x; t) = t
1
L(t)#(n; t); (8)
where L(t) is the crack lengthscale which is to be dened further for relevant fracture propagation regimes, and
e(t) = L
2
(t)V (t) (9)
is a small dimensionless parameter with the meaning of a characteristic crack aspect ratio. Using the alterna-
tive scaling for the opening and the velocity
X(n; t) = X(n; t)=c(t);

#(n; t) = #(n; t)=c(t);
and scaling Eqs. (7)(9), governing Eqs. (2) and (3), (5) and (6) can be written as follows:
uid mass:
t
_
V
V
_
1
n
Xdn
t
_
L
L
nX W
T
= X

#;
_
1
0
Xdn =
1
2c
2
; (10)
uid momentum:
(
q
c
2

#
t
_
L
L
1
n

#
o

#
on
_ _
1
o

#
on
U
T
_ _
(
m

#
X
2
=
oP
on
; (11)
LEFM:
X(n; t) = L
1
P(n; t); lim
n1
(1 n)
1=2
X = (
k
c
1=2
. (12)
The overdot in above equations corresponds to the time derivative, o/ot, and elasticity operator L
1
in Eq. (12)
is dened by
L
1
P(n; t) =
4
p
_
1
0
G(n; n
/
)P(n
/
; t)dn
/
. (13)
with the kernel G(n, n
/
) given by Eq. (5)
b
. The terms W
T
and U
T
are time-transient parts in the continuity Eq.
(10) and momentum Eq. (11), respectively,
W
T
= t
_
1
n
_
X
_ c
c
X n
oX
on
_ _ _ _
dn; U
T
=
t _ c
c
1
n

#
o

#
on
_ _

t
_

#
. (14)
Three dimensionless groups (
k
, (
m
, and (
q
appearing in Eqs. (10)(12) are dened as
(
k
=
K
/
E
/
L
3=2
V
; (
m
=
l
/
E
/
L
6
tV
3
; (
q
=
q
E
/
L
4
t
2
V
(15)
and quantify the relative eects of solid toughness, uid viscosity, and uid inertia, respectively. Formally, an
algebraic constraint of the form (
a
k
(
b
m
(
c
q
= 1 can be applied to these groups to x the fracture lengthscale L in
Eqs. (7)(9). Assuming injection power law of time, V ~ t
a
, the fracture lengthscale L and the nal form of the
dimensionless groups resulting from the above constraint are respective power laws of time. Consequently,
t
_
V =V and t
_
L=L in continuity Eq. (10)
a
and momentum Eq. (11) are the corresponding constant exponents,
and the time derivative operator to()/ot in Eqs. (10), (11) and (14) can be replaced by
t
o
ot
=
t
_
(
k
(
k
_ _
(
k
o
o(
k

t
_
(
m
(
m
_ _
(
m
o
o(
m

t
_
(
q
(
q
_ _
(
q
o
o(
q
; (16)
where coecients in the brackets are also constant exponents in the time power laws for (
k
, (
m
, and (
q
, respec-
tively. Consequently, the time dependence of the normalized solution of Eqs. (10)(12) can be replaced by that
of dimensionless groups, resulting, e.g., for the opening X = X(n; (
k
; (
m
; (
q
).
460 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
The three dierent physically motivated scalings [Eqs. (7)(9)] (corresponding to the three dierent deni-
tions of fracture lengthscale L) of the governing Eqs. (10)(12) result from setting one of the three dimension-
less groups [Eq. (15)] to unity. The lengthscale L, and the expressions of the dimensionless groups in these
three scalings, identied as the toughness- ((
k
= 1), the viscosity- ((
m
= 1), and the inertia- ((
q
= 1) scalings,
are presented in Table 1. The remaining two (non-unity) dimensionless groups in each scaling then dene the
parameters governing the solution of Eqs. (10)(12). These parameters can have the meaning of either non-
dimensionalized toughness (/s), or viscosity (/s), or inertia eects (1s), see Table 1. The toughness- and
the viscosity scalings are naturally related to the asymptotic cases where dissipation in the solid due to frac-
turing prevails over the dissipation in the viscous uid (viscosity is irrelevant, (
m
= /
k
= 0) and vice a versa
(toughness is irrelevant, (
k
= /
m
= 0), respectively [24]. Corresponding number (
q
(1
k
or 1
m
) quanties the
eect of inertia in these two scalings. The inertia-scaling can be naturally related to the asymptotic case where
inertia prevails over the viscous shear stresses (viscosity is irrelevant, (
m
= /
q
= 0). Parameter (
k
= /
q
in the
latter case quanties the eect of toughness in the inertia-dominated solution.
Solution of the governing Eqs. (10)(12) in either one of the three scalings in Table 1 depends on two
parameters only (with the remaining dimensionless group equal to unity). Governing parameters in the three
scalings are interrelated as follows:
/
k
= /
4
m
; 1
k
= /
2=3
k
1
m
; /
q
= /
k
/
4
q
(17)
while the relations between dimensionless eld variables X, P,

#, and c in dierent scalings can be readily
established from scaling Eqs. (7)(9) and the following relations between the three lengthscales L
k,m,q
:
L
m
L
k
= /
2=3
m
;
L
q
L
k
= /
2=3
q
. (18)
Examination of the toughness and the viscosity scalings shows that in the case of injection with a constant rate
Q
0
(fracture volume is V = Q
0
t), the viscosity /
k
and toughness /
m
parameters are constants while the cor-
responding inertia parameters are negative power law of time (1
k;m
~ t
1=3
). Consequently, inertia eects may
be important at early stages of fracture propagation, but vanish with time. The corresponding large-time
zero-inertia (1
k;m
= 0) asymptotic solution is self-similar, as originally observed by Spence and Sharp [6], since
the remaining parameter /
k
or /
m
is time-independent (the transient part of the uid ux in uid continuity
Eq. (10) is zero, W
T
= 0). This solution is characterized by the fracture length varying as t
2/3
. On the other
hand, early-time zero-toughness inertia-dominated asymptotic solution in inertia-scaling, /
q
= /
q
= 0, is
also self-similar, with the fracture length varying as t
3/4
.
Huang et al. [17] have considered a particular case of increasing injection rate,
_
V ~ t
1=5
, which in the inertia
scaling [Table 1] implies a constant value for the toughness parameter /
q
and a positive time power law for the
viscosity parameter, /
q
~ t
1=5
. Their self-similar solution for inviscid uid (/
q
= 0) characterized by the
fracture length varying as t
4/5
, corresponds, therefore, to the early-time solution. In the case when injection
rate
_
V increases with time faster than t
1/3
, eects of both toughness and viscosity on the fracture propagation
decrease with time (see the inertia scaling in Table 1) as the fracture evolves towards the inertia-dominated
regime.
Examination of the toughness, the viscosity, and the inertia scalings during the shut-in (when the injection
rate is suddenly reduced to zero and the injected volume remaining constant thereafter),
_
V = 0, shows that
Table 1
Quantities corresponding to toughness (k), viscosity (m) and inertia (q) scalings
Scaling L (
k
(
m
(
q
(k)
L
k
=
E
/
V
K
/
_ _
2=3
1
/
k
=
l
/
V
E
/
t
E
/
K
/
_ _
4
1
k
=
qE
/5=3
V
5=3
K
/8=3
t
2
(m) L
m
=
E
/
V
3
t
l
/
_ _
1=6
/
m
=
K
/
E
/
E
/
t
l
/
V
_ _
1=4
1 1
m
=
qV
l
/2=3
E
/1=3
t
4=3
(q) L
q
=
E
/
Vt
2
q
_ _
1=4
/
q
=
K
/
E
/
E
/
t
2
qV
5=3
_ _
3=8
/
q
=
l
/
E
/1=2
t
2
q
3=2
V
3=2
1
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 461
/
k
~ t
1
, 1
k
~ t
2
; /
m
~ t
1=4
, 1
m
~ t
4=3
; and /
q
~ t
3=4
, /
q
~ t
2
, respectively. Thus, the shut-in solution
evolves towards the zero-inertia toughness-dominated asymptotic limit (/
k
= 1
k
= 0), while the early-time
shut-in regime has to be determined by the corresponding regime at the end of the injection phase (prior to
the shut-in) given the continuity of the governing parameters between the injection and the shut-in.
Following the above discussion, the criterion for negligible uid inertia eect on hydraulic fracture pro-
pagation can be formulated as the smallness of appropriate inertia parameter
1
k
=
qE
/5=3
V
5=3
K
/8=3
t
2
1 when /
k
=
l
/
V
E
/
t
E
/
K
/
_ _
4
K1; (19)
1
m
=
qV
l
/2=3
E
/1=3
t
4=3
1 when /
k
J1. (20)
The duality of the above criterion reects a simple observation that the inertia eects are quantied by two
dierent parameters (1
k
vs. 1
m
) in the two extremities of fracture propagation dominated by the toughness
(/
k
1) and the viscosity (/
k
1), respectively. Let us consider a parameter set consistent with the appli-
cation of hydraulic fracturing to the well-stimulation in petroleum industry: q = 10
3
kg/m
3
, l = 10
7
MPa s
(100 cp), E
/
= 10 GPa, K
Ic
= 1 MPa m
1/2
and a constant injection rate value Q
0
= 10
4
m
2
s
1
. For the above
set at t = 1 min and at t = 1 h the fracture lengthscales are L
k
~ L
m
~ 7 m and L
k
~ L
m
~ 100 m, respectively,
and the inertia parameters are 1
k
~ 1
m
~ 10
5
and 1
k
~ 1
m
~ 10
6
, respectively. Clearly, the eect of inertia
can be neglected in this case. Furthermore, the dependence of inertia numbers on pumping time is weak
(1
k;m
~ t
1=3
) and injection rates as high as Q
0
~ 0.1 m
2
s
1
would be required for a sizable inertia eect
(1
k;m
~ 1) for the propagation time t > 1 min. Even though, out of the practical range in traditional hydraulic
fracturing treatments, high values of the injection rate are conceivable over a short time period in an explosive-
generated fracture.
The main focus of the following analysis is on the case where the uid inertia can be neglected (1
k;m
= 0)
and the eect of the uid viscosity on the fracture propagation is small compared to the eect of the material
toughness (/
k
1 and /
m
1). We acknowledge, however, that the perturbation method, which is devel-
oped in the forthcoming Section 6.1 in order to study small viscosity eects in the inertia-less solution, is
equally applicable to study other small eects, e.g., the eects of the uid inertia (see Appendix C).
4. Equations in toughness scaling with no inertia
Unless stated otherwise, we assume the case of the inertia-less uid ow in a fracture [Eqs. (19) and (20)],
which results in the lubrication theory [Eq. (11) with (
q
= 0 or Eq. (4)]. Solution in the toughness scaling then
depends on a single parameter, the dimensionless viscosity /
k
. Omitting hereafter the scaling index k for
brevity, solution in the toughness scaling
T(n; /) = X(n; /); P(n; /);

#(n; /); c(/) (21)
is governed by the reduced form of uid mass and momentum Eqs. (10) and (11)
t
_
V
V
_
1
n
Xdn
2
3
nX
_ _
W
T
=
X
3
/
oP
on
;
_
1
0
Xdn =
1
2c
2
(22)
and the unchanged LEFM Eq. (12)
X(n; /) = L
1
P(n; /); lim
n1
(1 n)
1=2
X = c
1=2
. (23)
In the continuity Eq. (22)
a
, reduced Poiseuille momentum Eq. (11),

# =
X
2
/
oP
on
, has been used, and elasticity
operator L
1
in Eq. (23)
a
is dened by Eq. (13). The expression for the transient part W
T
[Eq. (14)] of the uid
ux is given by
W
T
= t
_
/
_
1
n
oX
o/

1
c
dc
d/
X n
oX
on
_ _ _ _
dn. (24)
462 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
Consider a uid injection law in the form
Q = Q
0
[H(t) H(t t
+
)[ (25)
(where H() is the Heavyside function) corresponding to a constant rate
_
V = Q
0
from t = 0 to a certain time t
*
,
followed by the shut-in. The form of parameters in the toughness scaling [Table 1] and of the lubrication Eq.
(22) can be further particularized for the two parts of the injection history [Eq. (25)] as follows.
During the injection phase at a constant rate Q
0
(0 < t < t
*
), the crack lengthscale L
k
and the dimensionless
viscosity / reduce to
L
k
=
E
/
Q
0
t
K
/
_ _
2=3
; /=
l
/
Q
0
E
/
E
/
K
/
_ _
4
; 0 < t < t
+
. (26)
Since / is a constant, the transient term W
T
in Eq. (22)
a
vanishes and Eqs. (22) assume a self-similar form
/
_
1
n
Xdn
2
3
nX
_ _
= X
3 dP
dn
;
_
1
0
Xdn =
1
2c
2
. (27)
This is the self-similarity originally discovered by Spence and Sharp [6].
During the shut-in (t > t
*
), the injection volume remains constant, V(t) = V
*
= Q
0
t
*
, and the crack length-
scale L
k
and the dimensionless viscosity / reduce to
L
k
=
E
/
V
+
K
/
_ _
2=3
; /=
l
/
V
+
E
/
t
E
/
K
/
_ _
4
= /
+
t
+
t
; t > t
+
; (28)
where /
+
= l
/
Q
0
E
/3
=K
/4
is the constant value of the dimensionless viscosity at t = t
*
, according to Eq. (26)
b
,
and V
*
= Q
0
t
*
is the constant value of the crack volume during the shut-in. Since the term multiplied by
t
_
V
V
in
Eqs. (22) is zero for t > t
*
, Eqs. (22) with (24) become
t > t
+
:
_
1
n
oX
o/
1

1
c
dc
d/
1
X n
oX
on
_ _ _ _
dn = X
3 oP
on
;
_
1
0
Xdn =
1
2c
2
; (29)
where /
1
plays a role of dimensionless time [Eq. (28)
b
]. Note that, even though injection rate is discontin-
uous at t
*
, the scaling lengthscale L
k
is continuous, as well as the normalized solution for the opening X, the
net pressure P, and the crack half-length c.
The dimensionless viscosity / characterizes the relative importance of the uid viscous properties l
/
and
the solid toughness K
/
for hydraulic fracture propagation. In this paper we focus on the large toughness (or,
alternatively, small viscosity) solution for a uid-driven fracture. The large toughness solution is sought as a
series expansion in terms of the viscosity /, with the rst term of the expansion, O(1), corresponding to the
zero-viscosity solution (/= 0) and the subsequent terms (/
1
, /
2
, etc.) providing the next-order viscosity
corrections. Under conditions when the viscosity correction is small, the fracture is said to propagate in the
toughness-dominated regime, which corresponds to the zero-viscosity solution.
5. Zero-viscosity solution
The zero-viscosity solution T
0
(n) = T(n; 0) is independent of the injection law. Eqs. (22) and (23) with
/= 0 yield
0 = X
3
0
dP
0
dn
; c
2
0
= 2
_
1
0
X
0
dg; (30)
X
0
= L
1
P
0
; lim
n1
(1 n)
1=2
X
0
= c
1=2
0
. (31)
Solution of Eqs. (30) and (31) corresponds to uniform net pressure along the crack and the corresponding clas-
sical elliptic crack opening
X
0
=
p
1=3
2

1 n
2
_
; P
0
=
p
1=3
8
0:1831; c
0
=
2
p
2=3
0:9324. (32)
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 463
6. Large-toughness solution: Constant injection rate
6.1. Series expansion in /
In this case, the viscosity /is a constant and the corresponding solution T(n; /) is self-similar (i.e., time-
independent). We look for a series expansion of the self-similar solution [Eq. (21)] of Eqs. (27) and (23) in the
small parameter /,
T(n; /) =

j=0
/
j
T
j
(n) with T
j
(n) = X
j
(n); P
j
(n); c
j
. (33)
(The opening in the original scaling, X = cX can be obtained as the product of the respective expansions in Eq.
(33).) The rst n + 1 terms of the above series expansion dene an approximate solution of order n. The zero-
order solution T
0
(n), i.e., the 1st term in the expansion is given by the zero-viscosity solution [Eq. (32)]. The
rst-order solution T
0
(n) /T
1
(n) provides the correction of order / to the zero-viscosity solution and is
referred to as the small viscosity (large toughness) solution.
Substituting Eq. (33) in the governing Eqs. (27) and (23), and retaining terms of order / yield
_
1
n
X
0
dn
2
3
nX
0
= 3X
2
0
X
1
dP
0
dn
X
3
0
dP
1
dn
; c
1
= c
3
0
_
1
0
X
1
dg; (34)
X
1
= L
1
P
1
; lim
n1
(1 n)
1=2
X
1
=
1
2
c
1
c
3=2
0
. (35)
To integrate Eqs. (34) and (35), we rst substitute the zero-viscosity solution [Eq. (32)] into Eq. (34)
a
and inte-
grate to nd
P
1
= P
1
(0) DP
1
(n); DP
1
(n) =
8
3p
2=3
ln

1 n
2
_

3
4
ncos
1
n

1 n
2
_
_ _
; (36)
where the pressure correction at the fracture inlet, P
1
(0), is yet to be determined. Given Eq. (36), the corre-
sponding X
1
(n) [Eq. (35)
a
], is as follows:
X
1
= 4P
1
(0)

1 n
2
_
DX
1
(n); DX
1
(n) = L
1
DP
1
(n). (37)
Closed-form expression for DX
1
(n) is derived in Appendix A, see Eq. (A.12).
Using Eq. (37) with (A.12), we can evaluate the tip asymptote of X
1
(n) and c
1
[Eq. (34)
b
], as
lim
n1
(1 n)
1=2
X
1
(n) = 4

2
_
P
1
(0) Dj
1
; c
1
=
8
p
P
1
(0) Dc
1
; (38)
where the constants Dj
1
and Dc
1
are given by
Dj
1
=
8

2
_
p
_
1
0
DP
1
dn

1 n
2
_ =
56

2
_
ln 2
3p
2=3
;
Dc
1
= c
3
0
_
1
0
DX
1
dn = c
3
0
_
1
0
4

1 n
2
_
DP
1
dn =
8(8 ln 2 1)
3p
5=3
.
Comparison of Eqs. (35)
b
and (38)
a
using Eq. (38)
b
yields
P
1
(0) =
1 48 ln 2
9p
2=3
. (39)
Combination of Eqs. (36), (37), (A.12), (38
b
), and (39), yields the next-order term T
1
(n) in the series expansion
[Eq. (33)]
464 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
X
1
=
8
3p
2=3
2p 4nsin
1
n
5
6
ln 2
_ _
1 n
2
_

3
2
ln
1

1 n
2
_
_ _
1

1n
2
_
1

1 n
2
_
_ _
1

1n
2
_
_

_
_

_
_
_
_
_
_
_
_
_
; (40)
P
1
=
8
3p
2=3
1
24
ln 4

1 n
2
_ _ _

3
4
ncos
1
n

1 n
2
_
_ _
; (41)
c
1
=
32(1 6 ln 2)
9p
5=3
2:7220. (42)
Proles of opening and pressure for the zero-viscosity solution [Eq. (32)] and the rst-order terms [Eqs. (40)
(42)] are shown in Fig. 2.
6.2. Discussion
Similarly to the rst-order term above, we can work out terms in the expansion [Eq. (33)] to any given
order, O(/
j
), j P2. These terms and their integrals have to be implemented numerically in view of their
increasing complexity with increasing term-order. The results for the coecients of the high-order expansion
terms (up to O(/
10
)) for the crack half-length c, the net pressure P(0), and the opening X(0) at the inlet are
summarized in Table 2.
The plot of the zero-order O(1), the rst-order O(/) and the higher order O(/
10
) small viscosity solutions
for the crack half-length c [dashed lines in Fig. 3] in comparison to the nite viscosity numerical solution [15]
(open circles) show that the small viscosity expansion diverges at /= O(0:01). Using standard methods of
analysis and improvement of diverging series [25], we can establish that the small viscosity series diverge
due to a nearest (non-physical) singularity on the negative real axis of the parameter /, /= /
si
with
/
si
0:0333, of the form c ~ const (//
si
)
1=2
, see Appendix B. This suggests an Euler transformation,
or a reformulation of the small viscosity series in terms of the following small parameter
d(/) =
/

1 /=/
si
_ . (43)
This transformation eliminates the singularity and increases the radius of series convergence (improves the ser-
ies). The series in terms of d(/) are obtained by the substitution of the inverse of Eq. (43), /(d), into the
viscosity expansion [Eq. (33)] and expanding the result into the Tailor series in d. Since d(/) is equal to
the rst order to /, the improved rst-order small viscosity solution is of particularly simple form,
T
0
(n) d(/)T
1
(n).
The plot of the O(d) and the O(d
10
) solutions [solid lines in Fig. 3] show at least an order of magnitude
increase of the radius of convergence of the small viscosity expansion. The eventual divergence of the
improved series may indicate the emergence of a new state of the solution inuenced by the viscosity to an
2
-2
-1
0
1
2
4
6
0 0.2 0.4 0.6 0.8 1
0
0 0.2 0.4 0.6 0.8 1

1
0
1
0
(a) (b)
Fig. 2. (Constant injection rate, zero inertia) Zero- and rst-order viscosity terms of normalized opening, (a), and net-pressure, (b), in the
small viscosity solution.
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 465
extent which cannot be captured by corrections to the otherwise toughness-dominated state. This new state
has to be intimately related to the existence of an intermediate tip asymptote dened by viscous dissipation
[11], and characterized by the opening varying as (1 n)
2/3
at some (small) distance away from the tip
n = 1, which cannot be correctly captured by the small-viscosity solution with the near tip opening behavior
of the form (1 n)
1/2
.
The improved rst-order solution for the net pressure, P = P
0
d(/)P
1
and the opening in the original
scaling X = (c
0
d(/)c
1
)(X
0
d(/)X
1
) is shown in Fig. 4 for various values of the viscosity /=
0; 0:01; 0:03; 0:1. The O(d) solution agrees well with the numerical solution [15] (shown by open circles) with
some departure observed at a higher viscosity value /= 0:1. (Note that the O(d) approximation of the crack
length c [Fig. 3] is apparently more accurate than the O(d) approximation for the opening and the net pres-
sure.) Since the rst-order solution depends linearly on d(/) = /O(/
2
), the increase in the viscosity /
from 0 to 0.1 entails a signicant departure from the zero-viscosity limit (Fig. 4). On the other hand, the solu-
tion converges rapidly towards the zero-viscosity limit with increasing toughness / = /
1=4
[Eq. (17)
a
].
Owing to the logarithmic tip singularity of the viscosity correction to the net pressure, d(/)P
1
(n), the zero-
viscosity approximation breaks down near the fracture tip. Indeed, the approximation error introduced by the
zero-viscosity solution for the net pressure is O(d(/)) away from the tip, 1 n = O(1), and O(1) or greater in
the near tip region dened by 1 n = O exp
3p
32
/
1
_ _ _
. Consequently, the viscosity correction to the net
pressure has to be accounted for in the solution regardless of the smallness of /in order to accurately capture
Table 2
Numerical values of coecients of the O(/
j
) terms (j = 0, 1, . . . , 10) in the expansion [Eq. (33)] of crack half-length c, net pressure P(0)
and normalized opening X(0) at the inlet
c
j
0.04
j
P
j
(0) 0.04
j
X
j
(0) 0:04
j
j = 0 +0.93239 +0.18307 +0.73229
j = 1 0.10888 +0.07101 +0.30548
j = 2 +0.05961 0.03171 0.07971
j = 3 0.04528 +0.02359 +0.05544
j = 4 +0.04003 0.02099 0.04726
j = 5 0.03864 +0.02049 +0.04479
j = 6 +0.03951 0.02120 0.04532
j = 7 0.04206 +0.02282 +0.04792
j = 8 +0.04613 0.02527 0.05234
j = 9 0.05177 +0.02860 +0.05857
j = 10 +0.05917 0.03294 0.06681
0.001 0.01 0.1 1
0.5
0.6
0.7
0.8
0.9
1.0

Fig. 3. (Constant injection rate, zero inertia) Zero-viscosity O(1) solution and small viscosity solutions for the crack half-length: the
original series expansions [Eq. (33)] to O(/) and O(/
10
) (dashed lies) and the improved series expansions to O(d) and O(d
10
) (solid lines).
Open circles show nite toughness (viscosity) numerical solution (after Adachi [15]).
466 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
the near tip behavior. Note, however, that the extent of the latter near tip region, where the zero-viscosity pres-
sure solution P
0
(n) is not an accurate approximation, is exponentially small.
The zero-viscosity solution governs fracture propagation in the toughness-dominated regime, whilst the
small viscosity solution provides a viscosity correction to the latter solution, and, therefore, can be used to
estimate the parametric range of the toughness-dominated regime. Indeed, the threshold value of viscosity
/
0
3:4 10
3
for the toughness-dominated regime (/< /
0
) can be obtained by requiring a 1% maximum
deviation in the dimensionless fracture length from the zero-viscosity solution, i.e., /
0
= 0:01[c
0
=c
1
[
0:0034. The corresponding threshold value of toughness /

= /
1=4
0
for the toughness-dominated regime
(/ > /

) is then estimated as /

4:13.
6.2.1. Remark on small inertia eects on zero-viscosity solution
The perturbation method suggested in Section 6.1 to study the eect of small viscosity on the inertia-less
propagation of a plane-strain uid-driven fracture can be applied (i) to other one-dimensional fracture geom-
etries (e.g., an edge crack or a radial crack) and/or (ii) to analyze other small eects in the uid ow in a prop-
agating fracture (e.g., inertia).
Indeed, in the case of a dierent one-dimensional fracture geometry, the method is directly applicable given
the appropriate form of kernel G [Eq. (13)] in the elasticity integral operator L
1
, see Eqs. (31)
a
and (35)
a
;
and, in the case of a radial fracture [10], the appropriate form of the toughness scaling and of the lubrication
equations.
1
On the other hand, the analysis using the method of Section 6.1 of the eect of small uid inertia on the
propagation of a plane-strain fracture driven by inviscid uid is given in Appendix C. The small inertia solu-
tion is obtained in the toughness scaling in the form
T =

j=0
(1(t))
j
T
j
(n) with T
j
(n) = X
j
(n); P
j
(n); c
j
; (44)
where T
0
(n) is the zero-viscosity, zero-inertia solution [Eq. (32)] and subsequent terms are the inertia correc-
tions. The zero-order (zero-inertia) and the next-order small inertia solutions up to the O(1
5
) for the fracture
dimensionless half-length c(1) are shown in Fig. 5.
The O(1
5
) small inertia solution for the dimensionless opening X and the net pressure P is shown in Fig. 6
for various values of inertia parameter, 1 = 0; 0:1; 0:2; 0:3; 0:4; 0:5. It is interesting to point out that the net
pressure increases in the direction of the ow from the injection point to the fracture tip. This result, also
reported by Huang et al. [17] for a special case of increasing injection rate with time, can be attributed to
the Bernoullis eect for the ow of inviscid uid with the decreasing velocity in the direction of the ow.
For a ow of a viscous uid with negligible inertia, the net pressure decreases in the direction of the ow
due to the viscous dissipation, see, e.g., Fig. 4(b). Conceivably, if eects of viscosity and inertia are compara-
ble, the net pressure may develop a non-monotonic prole with the maximum in the interior of the crack.

0.0
0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0


0.2
0.4
0.6
0.8

0.12
0.15
0.18
0.21
0.24
0.27
0.30
=0
=0.1
(a) (b)
0
0.01
0.03
0.1
Fig. 4. (Constant injection rate, zero inertia) Dimensionless opening X = cX and net pressure P in the improved rst-order small
viscosity solution for various values of /.
1
The scaling and the lubrication equations for the case of an edge plane-strain crack are identical to that of a Griths crack.
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 467
Since the net pressure in the zero-viscosity solution has the minimum at the injection point, the crack may
tend to develop a tear-drop shape for larger values of the inertia parameter 1, as shown in Fig. 6(a).
7. Large-toughness solution: Shut-in
In this section we construct a solution for a uid-driven fracture during the shut-in (t Pt
*
) which follows
the injection phase at a constant rate Q
0
(t < t
*
) described by the rst-order large-toughness solution developed
in Section 6.
7.1. Nature of the solution
Let us dene a dimensionless time s starting with the shut-in instant, i.e., s(t
*
) = 0,
s =
1
/

1
/
+
=
1
/
+
t
t
+
1
_ _
; (45)
where /(t) = /
+
t
+
=t is the dimensionless viscosity during the shut-in (t Pt
*
), see Eq. (28), and /
+
is the
constant value of dimensionless viscosity prior to the shut-in (t 6 t
*
), see Eq. (26). In view of Eq. (45), it is
convenient to dene the solution in terms of the shut-in time s (instead of /), as
T
s
(n; s) = T(n; /(s)). (46)

0.90
0.95
1.00
1.05
1.10
1.15
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Fig. 5. (Constant injection rate, zero viscosity) Small inertia solutions [Eq. (33)] for crack half-length to O(1
n
); c =

n
j=0
1
j
c
j
, with
n = 0, 1, . . . , 5.

0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
0.2
0.4
0.6
0.8

0.00
0.05
0.10
0.15
0.20
0.25
=0.5
=0
(b) (a)
=0.5
=0
0.4
0.2
0.1
0.3
Fig. 6. (Constant injection rate, zero viscosity) The O(1
5
) small inertia solution for dimensionless opening X = cX, (a), and net pressure
P, (b), for various values of 1.
468 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
Prior to the shut-in (s 6 0), the solution is given by the self-similar large-toughness solution in the constant
injection rate case [Eq. (33)], with /= /
+
1,
T
s
(n; s 6 0) = T(n; /
+
) = T
0
(n) /
+
T
1
(n) ; (47)
where the zero-viscosity solution T
0
(n) and the next-order viscosity correction T
1
(n), are given by Eqs. (32)
and (40)(42), respectively. Eq. (47) then provides the initial condition for the shut-in solution T
s
(n; s P0),
which is governed by Eq. (29) with o(/
1
) = os, (45), and (23). The self-similar zero-viscosity solution T
0
(n)
satises the shut-in governing equations, but contradicts the initial condition at s = 0 [Eq. (47)], which diers
from T
0
(n) by terms of order /
+
. Consequently, the shut-in solution has to evolve from the initial viscosity-
dependent state [Eq. (47)] towards the zero-viscosity large-time limit. This evolution of the solution eectively
corresponds to a time-removal of the initial viscosity term /
+
T
1
(n) in Eq. (47). Acknowledging the shut-in
time scaling [Eq. (45)] and the large-toughness condition, /
+
1, evolution of the fracture to the toughness-
dominated regime given by the zero-viscosity solution T
0
(n) will take place over a small-time interval,
t t
*
t
*
, and can be viewed as a time boundary layer.
Based on the initial condition [Eq. (47)] and the smallness of /
+
, we can present the shut-in solution in the
form of a series expansion in /
+
T
s
(n; s P0) = T
0
(n) /
+
T
1s
(n; s) with T
1s
(n; 0) = T
1
(n). (48)
The termT
1s
(n; s) = X
1s
(n; s); P
1s
(n; s); c
1s
(s) is the time-dependent correction to the zero-viscosity solution
T
0
(n). The former vanishes for large time s, T
1s
(n; s ) = 0.
Substitution of the series expansion [Eq. (48)] with Eqs. (32) and (45) into Eqs. (29), and (23) while retaining
terms of order /
+
, allows us to obtain the governing equations for T
1s
(n; s) as follows:
_ c
1s
cos
1
n
4
p
_
1
n
_
X
1s
dn = 2(1 n
2
)
3=2
oP
1s
on
; c
1s
=
8
p
2
_
1
0
X
1s
dn; (49)
X
1s
= L
1
P
1s
; lim
n1
(1 n)
1=2
X
1s
=
pc
1s
2
5=2
; (50)
where the overdot corresponds to o/os.
Let us introduce the following representation for the opening X
1s
, pressure P
1s
, and crack length c
1s
terms:
X
1s
(n; s) = c
1s
(s)(n; s);
P
1s
(n; s) = c
1s
(s)T(n; s);
c
1s
(s) = c
1
exp
_
s
0
A(s)ds;
(51)
where A = _ c
1s
=c
1s
< 0 corresponds to the rate of exponential decay of the next-order viscosity correction c
1s
for the fracture length c
s
. In view of Eq. (51), and a new eld variable dened as
U = cos
1
n
4
p
_
1
n
dn; (52)
the uid ow Eq. (49) are reduced to
_
U = 2(1 n
2
)
3=2
oT
on
AU; U
[n=0
= 0; (53)
while the LEFM Eq. (50) become
= L
1
T; lim
n1
(1 n)
1=2
=
p
2
5=2
. (54)
Initial conditions for and T follow from Eq. (48), with Eq. (51) in the form
(n; 0) = X
1
(n)=c
1
; T(n; 0) = P
1
(n)=c
1
. (55)
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 469
The governing Eqs. (52)(54) in scaling [Eqs. (51)], suggest a self-similar (i.e., time-independent) large-time
solution when
_
U in Eqs. (53) vanishes and the exponent A(s) tends to a constant value A() < 0. This self-
similar solution has the meaning of an intermediate large-time asymptote, corresponding to the time when
the details of the imposed initial conditions [Eqs. (55)], become irrelevant. This large-time solution corre-
sponds to an exponential decay of the opening X
1s
, pressure P
1s
, and crack length c
1s
terms in time with con-
stant exponential rate A() < 0, see Eqs. (51).
7.2. Method of solution
According to Eq. (52),
_
U
[n=1
= 0 , and the asymptotic form of the lubrication Eq. (53)
a
near the tip can be
integrated in n to yield a pressure-term asymptote, T(n; s) =
1
4
A(s) ln(1 n), as n 1. The initial value
A(0) = 3p/(2(1 + 6ln2)) can then be obtained by evaluating the latter asymptote at s = 0 and applying
the pressure-term initial condition [Eq. (55)
b
] with P
1
(n) and c
1
given by Eqs. (41) and (42), respectively.
Finally, the pressure-term asymptote can be expressed in terms of the initial data and the time varying param-
eter A(s),
n 1: T(n; s) =
A(s)
A(0)
T(n; 0). (56)
Based on the asymptotic, Eq. (56), and initial value, Eqs. (55), considerations, we can represent the solution
for T and as a sum of (logarithmically) singular and regular parts as follows:
T =
A(s)
A(0)
T(n; 0) dT(n; s); =
A(s)
A(0)
(n; 0) d(n; s). (57)
Functions dT(n; s) and d(n; s) are regular in n [0, 1] with initial condition following from Eq. (55):
dT(n; 0) = d(n; 0) = 0. (58)
Continuity Eq. (53)
b
and elasticity Eqs. (54) yield
A(s)
A(0)
1 =
8
p
2
_
1
0
ddn; (59)
d = L
1
dT; lim
n1
(1 n)
1=2
d =
p
2
5=2
A(s)
A(0)
1
_ _
. (60)
Summarizing, via the representation [Eq. (57)], the solution for the pressure T, opening , and crack length
c
1s
reduces to the solution for the regular parts of the pressure and opening terms, dT(n; s) and d(n; s),
respectively, and the exponent A(s) given by Eqs. (52), (53), (59), (60) and initial condition [Eq. (58)]. The
large-time self-similar asymptote of the above solution is governed by the latter equations with
_
U = 0 in
the lubrication Eq. (53)
a
and the time-dependent exponent A(s) replaced by its limit value A().
Solution of Eqs. (52), (53), (59), (60), and (58) is sought via the numerical method of lines (see Liskovets [26]
for general discussion, and Nilson and Griths [27] and Garagash and Detournay [28] for applications of the
method to hydraulic fracture problems). This method approximates Eqs. (59) and (60) by a coupled system of
ordinary dierential equations (ODEs) governing the time evolution of the values of dT(n; s) at the xed
grid points over the space interval representing the crack. Details of the numerical method are given in
Appendix D.
7.3. Results and discussion
Calculations of the solution d(n; s); dT(n; s); A(s) of Eqs. (52), (53), (59), (60), and (58) have been car-
ried out using 26 grid points equally spaced along the half-length of the crack that include the inlet and the tip.
(Convergence of the numerical scheme has been evaluated by carrying out calculations with 11 grid points
which produced practically identical results.) The corresponding normalized next-order terms in opening,
= X
1s
=c
1s
, and pressure, T = P
1s
=c
1s
, [Eq. (57)], along the crack are shown in Fig. 7 for various times. This
470 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
gure shows that overall variation of and T with time s is marginal, and it is mainly driven by the step
change of the gradient of the pressure term T at the inlet, n = 0, from the nite value corresponding to the
initial conditions (constant non-zero injection rate) to zero value instantaneously upon the shut-in. Conse-
quently, the evolution of the next-order opening X
1s
= c
1s
, and pressure P
1s
= c
1s
T terms in the shut-in
solution expansion [Eq. (48)] is largely dened by the time evolution of the next-order term in the fracture
length, c
1s
.
Evolution of the length c
1s
and of its exponential rate A = _ c
1s
=c
1s
with time s is shown in Fig. 8 in a semi-
logarithmic scale. The next-order crack length term c
1s
evolves from the initial value c
1
[Eq. (42)] to zero at
large time, when the shut-in solution approaches its large-time zero-viscosity limit. The rate A of exponential
decay of c
1s
saturates at the large-time value A() 1.1 for s J s

= 0.33,
2
see Fig. 8. Consequently, for
s J s

, the next-order viscosity correction T


1s
(n; s) in the shut-in solution [Eq. (48)] is given by the interme-
diate large-time asymptote
X
1s
c
1s
(s)(n; ); P
1s
c
1s
(s)T(n; ); c
1s
c
1s
(s

)e
A()(ss

)
; (61)
-2
-3
-1
0
-3 -2 -1 1
10
log
0
-1.1
-1
-0.9
A
Fig. 8. (Shut-in, zero-inertia) Evolution of next-order crack length term c
1s
(s) and its exponential decay rate A = _ c
1s
=c
1s
.
0 0.2 0.4 0.6 0.8 1

-2
-1.5
-1
-0.5
0
-1
-0.5
0
0.5
1
Fig. 7. (Shut-in, zero-inertia) Distribution of pressure T = P
1s
=c
1s
and opening = X
1s
=c
1s
renormalized rst-order terms in the shut-in
solution expansion [Eq. (48)] along the crack for discrete time-set s =
1
/+
t
t+
1
_ _
= 0; 0:01; 0:05; 0:1; (time s = 0 corresponds to the
shut-in instant t = t
*
).
2
This time threshold corresponds to the 0.1% dierence between the solution A(s) and the asymptote A().
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 471
where c
1s
(s

) = c
1
exp
_
s

0
A(s)ds 1:923. Interestingly, the intermediate large-time asymptote for c
1s
[Eq.
(61)
c
] provides an excellent approximation for the whole shut-in time range, see dashed curve in Fig. 8.
Fig. 9 shows the distribution of the dimensionless opening X
s
and the net pressure P
s
along the fracture
length for /
+
= 0:01 at various times. The shut-in solution evolves from the initial state (s = 0) given by
the small viscosity-solution with /= 0:01 for the constant injection rate case towards the toughness-domi-
nated large-time limit with /= 0 [Fig. 4]. The solution is eectively given (within 1% accuracy) by the
large-time toughness-dominated limit for s J 3.08 (as inferred from the considerations for the crack length
[Eq. (61)
c
]), or, in physical time, t t
+
J3:08/
+
t
+
. Evolution of the normalized crack length
c
s
= c
0
/
+
c
1s
(s) during the shut-in can be trivially inferred from that of c
1s
(s), see Fig. 7(b).
Let us consider the evolution of fracture length in more detail. For an inviscid uid, /
+
= 0, the normal-
ized crack length is given by the zero-viscosity solution, c
s
= c
0
. Since the lengthscale of the fracture
L
k
= (E
/
V
*
/K
/
)
2/3
is constant during the shut-in [Eq. (28)
a
] (recall that L
k
~ t
2/3
[Eq. (26)
a
] during the constant
injection rate stage, t < t
*
), the fracture length remains unchanged during the shut-in, = L
k
c
0
. For small but
non-zero dimensionless viscosity /
+
, the dimensionless fracture length at the shut-in instant is given by
c
s
(s = 0) = c
0
/
+
c
1s
(s = 0), with c
1s
(s = 0) = c
1
2.72 [Eq. (42)]. Thus, time removal of the initial
viscosity length correction during the shut-in corresponds to further growth of the fracture by
Dc
s
= /
+
c
1
> 0. The dimensionless time interval during the shut-in, s
stop
, required for fracture dimension-
less length c
s
to grow within 1% of its terminal length c
0
, can be estimated from the intermediate large-time
asymptote [Eq. (61)] as s
stop
= 5:17 0:91 ln /
+
. For example for /
+
= 0:01, s
stop
1 and the corresponding
physical time interval Dt
stop
is given by Dt
stop
=t
+
= /
+
s
stop
0:01. The corresponding fracture length growth
during the shut-in as a fraction of its length
*
at the end of the injection phase (t = t
*
) is given by
D=
+
/
+
c
1
=c
0
2:9/
+
, or 2.9% for /
+
= 0:01. Similarly, the terminal net pressure drop at the fracture
inlet during the shut-in is Dp=p
+
/
+
P
1
(0)=P
0
(0) 9:6/
+
or 9.6% for /
+
= 0:01. Above points are illus-
trated in Fig. 10 which shows the injection and the shut-in phases of hydraulic fracture evolution in the small
viscosity solution for the fracture half-length and net pressure at the fracture inlet normalized by their respec-
tive values at the end of injection phase (t = t
*
) and various values of the viscosity parameter in the injection
phase /
+
= 0:003; 0:01; 0:03.
It is interesting to note that the fracture propagation during the shut-in can have much greater extent (than
estimated above) if fracture propagation during the preceding injection phase is not constrained to the large-
toughness parametric regime. Indeed, let us consider fracture propagation prior to the shut-in in the viscosity-
dominated regime /
+
1 (/
+
1) with fracture length at the shut-in instant given by (t
*
) = L
m
(t
*
)c
0
where
L
m
(t) = Q
1=2
0
E
/1=6
l
/1=6
t
2=3
and c
0
0.61524 [14]. The shut-in instant is followed by the fracture evolution dri-
ven by increasing dimensionless toughness / = /
+
(t=t
+
)
1=4
(or decreasing dimensionless viscosity
/= /
+
(t
+
=t)) towards the toughness-dominated regime. In the latter, the terminal fracture length is given
by
stop
= L
k
(t
*
)c
k0
with L
k
(t
*
) = (E
/
Q
0
t
*
/K
/
)
2/3
and c
0
0.9324. Resulting fracture length increase factor dur-
ing the shut-in
stop
=(t
+
) ~ /
2=3
+
= /
1=6
+
1 is exceedingly large. Even though, this large fracture growth
during the shut-in may be limited by the fracturing uid leak-o into the solid. Further investigation of the

= 0
= 0
=
=
0.2
0.4
0.6
0.8
0.17
0.15
0.19
0.21
0 0.2 0.4 0.6 0.8 1
0
0 0.2 0.4 0.6 0.8 1
(b) (a)
Fig. 9. (Shut-in, zero-inertia) Dimensionless opening X
s
= c
s
X
s
, (a), and net pressure P
s
, (b), along the crack at various times
s = {0, 0.01, 0.1, 0.5, 1, 2.5,} and initial value /
+
= 0:01 of the viscosity parameter.
472 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
above eects concerning fracture growth during the shut-in is currently underway using the method-of-lines
numerical method.
8. Closing discussion
In this paper we have derived the large-toughness asymptotic solutions for a plane-strain uid-driven frac-
ture propagating in an impermeable elastic solid. The part of this work which concerns the inertia-less prop-
agation complements the original numerical solution of the plane-strain hydraulic fracture problem by Spence
and Sharp [6] (further rened in [15]), the zero-toughness [14] and the small-toughness [11] asymptotic solu-
tions. Using the so-called toughness scaling of the problem, it is shown that the fracture propagation is gov-
erned by a dimensionless parameter with the meaning of a viscosity /=
l
/
V (t)
E
/
t
E
/
K
/
_ _
4
, or, alternatively, a
toughness / = /
1=4
; and additionally by a dimensionless inertia parameter 1 =
qE
/5=3
V (t)
5=3
K
/8=3
t
2
when eects of
the uid inertia are included. The latter inertia eects quantied by 1are negligible for usual applications with
possible exception of blasting (explosive-driven fractures). In the toughness-dominated regime, solution is
given by a similarity zero-viscosity zero-inertia solution whereby the dependence of the fracture half-length
, the opening w, and the net pressure p = p
f
r
0
on the problem parameters and time is prescribed solely
by the scaling [Eqs. (7)(9) and Table 1
(k)
]:
(t)
K
/
E
/
V (t)
_ _
2=3
c
0
;
w(x; t)
K
/
V
1=2
(t)
E
/
_ _
2=3
c
0
X
0
x
(t)
_ _
;
p(x; t)
K
/
E
/
V
1=4
(t)
_ _
4=3
E
/
P
0
;
(62)
where c
0
; X
0
(n); P
0
is the similarity solution [Eq. (32)]. The range of problem parameters and time for the
fracture to propagate in the toughness-dominated regime is constrained by the estimate on the dimensionless
viscosity, /1, and the dimensionless inertia, 1 1, and therefore, depends on the uid injection law.
The dependence of the solution on a small viscosity (small deviations from the toughness-dominated prop-
agation regime) is considered for two representative injection laws: (i) when the uid injection rate is a con-
stant; and (ii) during the shut-in.
Fig. 10. (Zero-inertia) Evolution of fracture half-length and net pressure p at the inlet normalized by the respective values at the shut-in
(t = t
*
) for various values of viscosity parameter /
+
in the injection phase.
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 473
8.1. Constant injection rate
In this case, we suggested a method of asymptotic series expansions which allows to study small eects of
uid viscosity or/and uid inertia on the otherwise zero-viscosity, zero-inertia solution of the toughness-dom-
inated regime of fracture propagation. The small-viscosity (/1) similarity expansion of the zero-inertia
solution allows to establish condition /< /
0
with /
0
3:4 10
3
(1% departure from the zero-viscosity
solution [Eq. (62)]) which denes a constraint on the solid parameters, uid viscosity, and uid injection rate
under which viscosity is irrelevant for uid-driven fracture propagation (toughness-dominated regime). The
latter threshold /
0
3:4 10
3
and the threshold /

4:16 for the viscosity-dominated regime


/> /

(1% departure from the zero-toughness solution, see [11]) provide the parametric range
/ (/
0
; /

) where both the toughness and the viscosity inuence the propagation of a hydraulic fracture.
Fig. 11 shows the variation of the dimensionless fracture length in the toughness scaling with the dimen-
sionless viscosity /, in
the small-toughness asymptotic solution,
3
c = /
1=6
(c
0
S(/
1=4
)c
1
)
with c
0
0.6152, c
1
0.1754,and S(/) = 0:1/
3:18
is the small-toughness (/ = /
1=4
) parameter [11];
and
the improved small-viscosity asymptotic solution,
c = c
0
d(/)c
1
with d(/) = /(1 /=/
si
)
1=2
and c
0
0.9324, c
1
2.7220, and /
si
0:0333. The nite toughness numerical solution of Adachi [15] is
also shown in this plot (open circles).
Remarkably, the range of viscosity in which the solution departs from the rst-order small-toughness solu-
tion and the improved rst-order small-viscosity solution is very narrow, 0:2 K/K0:4, see Fig. 11, i.e., these
two asymptotes, when combined, eectively approximate the solution in the complete range of the viscosity
parameter /.
0.001 0.01 0.1 1 10
c
r
a
c
k

l
e
n
g
t
h
,

(
E
'
V
(
t
)
/
K
'
)
-
2
/
3
0.4
0.6
0.8
1.0
3 4
viscosity, ( ) E K V t t

=
Fig. 11. (Zero-inertia) Dependence of dimensionless crack half-length on dimensionless viscosity /for a constant injection rate: (i) zero-
viscosity and zero-toughness [14] solutions are shown by dashed lines; (ii) improved small viscosity and small-toughness [11] asymptotic
solutions are shown by solid lines; and (iii) numerical nite toughness solution [15] is shown by open circles. Triangles show evolution of
the shut-in solution with diminishing viscosity parameter /= /
+
t+
t
for two initial values, /
+
= 0:01; 0:03 towards the toughness-
dominated regime, / 0.
3
Converted from the viscosity (m) to the toughness (k) scalings via c
k
= /
2=3
c
m
= /
1=6
c
m
[Section 3].
474 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
The small-inertia (1(t) 1) series expansion of the zero-viscosity solution allows to study the solution
departure from the inertia-less regime. In this case, the solution expansion is not self-similar (1(t) is a decreas-
ing function of time), but each coecient of the series is. The distribution of the net pressure in the small-iner-
tia, zero-viscosity solution shows an increase from the inlet to the tip (an opposite behavior to that of the
small-viscosity, zero-inertia solution where viscous dissipation in the ow reduces the pressure with the dis-
tance in the ow) due to the Bernoullis eect for the ow of inviscid uid in the crack [17]. Consequently,
with increasing inertia, the crack can develop a tear-drop shape with the maximum opening away from the
crack inlet [Fig. 6(a)].
The small-inertia expansion of the zero-viscosity solution and the small-viscosity expansion of the zero-
inertia solution can be simply combined to the rst order to obtain the O(/; 1) small viscosity, small inertia
solution. For the dimensionless net pressure we have
P(n; /; 1) = P
0
d(/)P
(l)
1
(n) 1P
(q)
1
(n) (63)
with P
(l)
1
(n) and P
(q)
1
(n) denote the expressions for the corresponding rst-order term P
1
(n) in [Eq. (41)] and
[Eqs. (C.9)(C.11)], respectively. Given that the viscosity term P
(l)
1
(n) and the inertia term P
(q)
1
(n) in [Eq. (63)]
decreases and increases in the direction of the ow, respectively. An investigation of Eq. (63) then shows that if
the eect of the inertia is comparable to or exceeds the eect of the viscosity (1 >
3
2
p
2=3
d(/)) the net pressure
develops a non-monotonic prole with the maximum in the interior of the crack.
8.2. Shut-in
During the shut-in (following an injection), the large-time limit corresponds to an immobile fracture given
by the zero-viscosity zero-inertia solution [Eq. (62)] for a constant injected volume V(t Pt
*
) = V(t
*
). Evolu-
tion of the large-toughness solution from the shut-in instant towards the large-time limit corresponds to a
slowing fracture and vanishing eect of the viscosity in time. Fig. 11 illustrates the convergence of the shut-
in solution for the crack length (shown by triangles) towards the toughness-dominated limit. The extent of
fracture propagation during the shut-in D as a fraction of the length
*
at the end of injection phase, is given
by D=
+
2:9/
+
. The latter is a small fraction for the considered case of a fracture initially propagating in
the large-toughness (small-viscosity) parametric regime, i.e., when /
+
is small. However, much greater extent
of the shut-in propagation is anticipated for fractures which propagation during the injection phase is char-
acterized by smaller values of the dimensionless toughness. This eect can be partly oset by a progressive
uid leak-o into the permeable rock in eld hydraulic fracturing applications, but can be expected to be
prominent for tight reservoir rocks or in the hydrofrac laboratory tests in impermeable materials (e.g., PMMA
[16]). The transient shut-in solution with the uid leak-o in arbitrary parametric regime (not conned to the
large-toughness case) would constitute an interesting problem for the future work.
Acknowledgements
Acknowledgement is made to the Donors of The Petroleum Research Fund, administered by the American
Chemical Society, for partial support of this research under Grant ACS-PRF 36729-G2. The author is grateful
to Emmanuel Detournay and Jose I. Adachi for interesting discussions and thorough review of the early ver-
sion of this manuscript, and to the anonymous reviewer for the comment which led to the improvement of the
small-viscosity series.
Appendix A. Closed-form expression for DX
1
(n), Eq. (37)
In this Appendix we provide details of the calculation of the integral expression for DX
1
(n), dened in Eq.
(37)
b
, with DP
1
(n) given in Eq. (36)
b
. This integral can be decomposed into two integrals corresponding to the
two terms in DP
1
(n) [Eq. (36)
b
],
DX
1
(n) =
8
3p
2=3
I
ln
(f)
3
4
I
cos
(f)
_ _
f =

1 n
2
_ _ _
(A:1)
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 475
with I
ln
(f) and I
cos
(f) given by
I
ln
(f) =
4
p
_
1
0
G(f; f
/
) ln f
/
d

1 f
/2
_
; (A:2)
I
cos
(f) =
4
p
_
1
0
G(f; f
/
)sin
1
f
/
df
/
; (A:3)
where the substitution f
/
=

1 n
/2
_
has been used. The kernel G(f; f
/
) is given by
G(f; f
/
) = G

1 f
2
_
;

1 f
/2
_ _ _
= ln
f f
/
f f
/

.
At rst, let us consider the integral I
ln
(f), which can be decomposed as
I
ln
(f) = I
(1)
ln
(f) I
(2)
ln
(f); (A:4)
I
(1)
ln
(f) =
4
p
_
1
0
ln
f f
/
f f
/

1 f
/2
_
f
/
df
/
; I
(2)
ln
(f) =
8f
p
_
1
0
ln f
/

1 f
/2
_
f
2
f
/2
df
/
.
Dierentiating integral I
(1)
ln
(f) with respect to the parameter f and then carrying out the integration in f
/
yields
dI
(1)
ln
=df = 4. Consequently, the integral can be expressed as I
(1)
ln
(f) = 4(1 f) I
(1)
ln
(1), where I
(1)
ln
(1) is a yet
unknown constant. Integration by parts and further simplications result in I
(1)
ln
(1) = 2p 4. Summarizing,
I
(1)
ln
(f) = 2p 4f. (A:5)
The second term I
(2)
ln
(f) in the expression for I
ln
(f) is rst integrated by parts which upon some algebra results
in
I
(2)
ln
(f) =
8
p
fJ
1

4
p

1 f
2
_
J
2
(f); (A:6)
J
1
=
_
1
0
sin
1
f
/
f
/
df
/
=
pln 4
4
; (A:7)
J
2
(f) =
_
1
0
ln
f f
/
f f
/
1 ff
/

1 f
2
_
1 f
/2
_
1 ff
/

1 f
2
_
1 f
/2
_

df
/
f
/
. (A:8)
Integral J
2
(f) upon the substitution f = sin a and f
/
= sin a
/
, and some algebra becomes

J
2
(a) = J
2
(sin a) =
_
p=2
0
cot a
/
ln
sin(a a
/
)
sin(a a
/
)

da
/
. (A:9)
Dierentiation of the integral

J
2
(a) with respect to the parameter a and further integration yields
d

J
2
(a)=da = p. Utilizing the known value of

J
2
at a = p/2,

J
2
(p=2) = 0, see Eq. (A.9), we nally obtain
J
2
(f) = pcos
1
f. (A:10)
Lastly, let us consider I
cos
(f). Dierentiating with respect to the parameter f, and carrying out the integration,
yields dI
cos
(f)/df = 2ln[4(1 f
2
)]. Utilizing condition I
cos
(0) = 0, we obtain
I
cos
(f) = 2 2(1 ln 2)f ln
(1 f)
1f
(1 f)
1f
_ _ _ _
. (A:11)
Substitution of Eqs. (A.11) and (A.4) with (A.5) and (A.6) into Eq. (A.1) yields the nal expression
DX
1
=
8
3p
2=3
2p 4nsin
1
n (1 7 ln 2)

1 n
2
_

3
2
ln
1

1 n
2
_
_ _
1

1n
2
_
(1

1 n
2
_
)
1

1n
2
_
_

_
_

_
_
_
_
_
_
_
_
_
. (A:12)
476 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
Appendix B. Improvement of small-viscosity series
The small-viscosity expansion [Eq. (33)] of the zero-inertia solution have the form of alternating series
[Table 2], which limits its radius of convergence to /= O(0:01). The latter is attributed to the nearest sin-
gularity of the series in the complex plane of the parameter /. The standard method of the analysis of diverg-
ing series using Domb-Sykes plot [25, p. 245] allows to nd both the location of the singularity and the
singularity strength. Consider, for example, the series for the fracture dimensionless half-length,
c(/) =

j=0
c
j
/
j
; (B:1)
where coecients c
j
, j = 1, . . . , 10, are found in Table 2. The Domb-Sykes plot for this series (plot of c
j
/c
j1
vs.
1/j) is shown in Fig. B.1. The reciprocal of the intercept at 1/j = 0 gives the location of the singularity on the
negative real axis of parameter /, /= /
si
with /
si
0:03333. The asymptotic slope S (2/
si
)
1
of the
plot is related to the strength a = /
si
S 1 1=2 of the algebraic singularity, c ~ const (//
si
)
a
as
//
si
.
Appendix C. Small inertia, zero-viscosity solution
Consider the case of a constant injection rate. Governing equations in the toughness scaling under condi-
tions of zero-viscosity, /
k
= 0, depend on a single evolution parameter 1
k
(t) with the meaning of inertia.
Omitting hereafter the scaling index k for brevity, the zero-viscosity solution (accounting for inertia eects)
in the toughness scaling [Eqs. (7)(9) and Table 1
(k)
]
T(n; 1) = X(n; 1); P(n; 1);

#(n; 1); c(1)
_ _
(C:1)
is governed by the reduced form of governing Eqs. (10)(12)
_
1
n
Xdn
2
3
nX W
T
= X

#;
_
1
0
Xdn =
1
2c
2
; (C:2)

1
3
1c
2

# 1
1
c

#
oc

#
o1
2
1
c
oc
o1
_ _
n

#
3
_ _
o

#
on
_ _
=
oP
on
; (C:3)
X(n; t) = L
1
P(n; t); lim
n1
(1 n)
1=2
X = c
1=2
; (C:4)
where identity t(o=ot) = (1=3)1(o=o1) was used (1 ~ t
1=3
[Table 1
(k)
]). The expression for the transient
part W
T
[Eq. (14)
a
] of the uid ux is given by
1/j
-40
0.0 0.1 0.2 0.3 0.4 0.5 0.6
-35
-30
-25
-20
-15
-10
-5
0
1
si
(2 ) S

1
si
33.3

1
j
j

Fig. B.1. Domb-Sykes plot for small viscosity expansion [Eq. (B.1)] of dimensionless fracture length c.
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 477
W
T
=
1
3
_
1
n
1
oX
o1

1
c
oc
o1
X n
oX
on
_ _ _ _
dn; (C:5)
while the expression for the transient part U
T
[Eq. (14)
b
] in the momentum equation has been used with the
above identity to arrive to Eq. (C.3).
The above system of equations explicitly depends on the evolving-in-time parameter 1 , and, therefore, its
general solution (except for the zero and the innite limits of 1) cannot be self-similar.
4
The limit of zero iner-
tia, 1 = 0, constitutes a self-similar limit given by Eq. (32). Since the eect of the uid inertia quantied by 1
decreases with time, the above limit corresponds to the large-time asymptote of a fracture driven by an inviscid
uid. The transition towards this self-similar limit can be studied using the perturbation method of Section 6.1.
Substituting small inertia expansion [Eq. (44)] into Eqs. (C.2)(C.5) and collecting terms of the same order
in 1 yield equations for the expansion terms to any given order. In constructing expansions for the governing
equations, the care should be exercised in evaluating expansions for the transient terms, e.g.,
1
c
oc
o1
=
c
1
c
0
1
2c
2
c
0

c
2
1
c
2
0
_ _
1
2
O(1
3
).
The contribution of transient terms in lubrication Eq. (C.3) is O(1
2
), consequently, the equations governing
the O(1) term in the expansion are particularly simple
1
X
0
_
1
n
X
0
dn
2
3
n =

#
0
; c
1
= c
3
0
_
1
0
X
1
dg; (C:6)

1
3
c
2
0

#
0
1 2
n

#
0
3
_ _
d

#
0
dn
_ _
=
dP
1
dn
; (C:7)
X
1
(n) = L
1
P
1
(n); lim
n1
(1 n)
1=2
X
1
=
1
2
c
1
c
3=2
0
. (C:8)
Substitution of the expression for

#
0
following from Eq. (C.6)
a
into Eq. (C.7) and integration yields the pres-
sure term prole
P
1
= P
1
(0) DP
1
(n) with (C:9)
DP
1
(n) =
1
6p
4=3
p
2
4

5n
2
3

6ncos
1
n

1 n
2
_
(1 2n
2
)(cos
1
n)
2
1 n
2
_ _
; (C:10)
where the inlet value P
1
(0) is unknown. Corresponding expression for the opening term X
1
= L
1
P
1
is of
the form (37). Following the treatment of Section 6.1 we can obtain the solution constants as follows:
P
1
(0) =
21p
2
13
216p
4=3
0:19546; (C:11)
c
1
=
4(6p
2
43)
27p
7=3
0:16621. (C:12)
Similarly, we obtain the higher order terms in the expansion (44). The integrals arising in the evaluation of the
elasticity and the lubrication equations for successive expansion terms increase in complexity, and, therefore,
are implemented numerically. The results for the terms in the inertia expansion of the zero-viscosity solution
up to the O(1
5
) for the crack length c, the inlet, P(0), and the tip, P(1), values of the net pressure, and the
inlet normalized opening X(0) are summarized in Table C.1. Proles of the opening and the pressure terms in
the inertia expansion of the zero-viscosity solution (44) are shown in Fig. C.1.
4
As discussed in Section 3, a self-similar solution exists when the injection rate is increasing with time as
_
V ~ t
1=5
. The treatment of this
or of any other power law injection case,
_
V ~ t
a
with a > 0, is identical to the treatment of a constant injection rate case considered here
when numerical coecients in Eqs. (C.2)
a
and (C.3) are modied accordingly.
478 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
Appendix D. Method of lines numerical approach
A general method-of-lines numerical approach [2628] is used to solve the system of coupled Eqs. (52), (53),
(59), (60), and (58) for the regular parts dT(n; s) and d(n; s) of the pressure and opening terms, respectively,
and crack length exponent A(s ). The method of lines as applied here essentially corresponds to a discretization
of the solution in space, but not in time, such that the governing equations reduce to a system of coupled
ODEs in time. This system governs the evolution of the values of the solution at a discrete set of points in
the space interval along the crack.
First, a set of xed grid points {n
i
}, i = 1, . . . , M with n
1
= 0 and n
M
= 1 is selected. (Even though a uniform
grid spacing is used in the computation of the nal results, the development below does not make this assump-
tion.) The regular function dT(n; s) is then approximated by a piecewise linear function dened by its values
dT
i
(s) at the grid points {n
i
}
dT(n; s) dT
i
(s)
dT
i1
(s) dT
i
(s)
n
i1
n
i
n n
i
( ); n [n
i
; n
i1
). (D:1)
The corresponding expression for d(n; s) is given by the elasticity integral [Eq. (60)
a
], computed explicitly for
the piecewise linear distribution of dT [Eq. (D.1)]
d(n; s) =

M1
i=1
[I
a
(n; n
/
)a
i
(s) I
b
(n; n
/
)b
i
(s)[
n
/
=n
i1
n
/
=n
i
. (D:2)
In the above expression a
i
(s) and b
i
(s) are given by
a
i
= dT
i

dT
i1
dT
i
n
i1
n
i
n
i
; b
i
=
dT
i1
dT
i
n
i1
n
i
; (D:3)
and the functions I
a,b
(n, n
/
) are dened as follows:
Table C.1
Numerical values of coecients of the O(1
j
) terms, j = 0, 1, . . . , 5, in the expansion [Eq. (44)] of crack half-length c, net pressure at the
inlet, P(0), net pressure at the tip, P(1), and normalized opening X(0) at the inlet
c
j
P
j
(0) P
j
(1) X
j
(0)
j = 0 0.93239 0.18307 0.18307 0.73229
j = 1 0.16621 0.19546 0.062952 0.376464
j = 2 0.154028 0.144179 0.0301009 0.259143
j = 3 0.200661 0.205116 0.028148 0.324525
j = 4 0.305301 0.381546 0.0280604 0.517134
j = 5 0.502836 0.816739 0.0168989 0.936821

-1.0
-0.8
-0.6
-0.4
-0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0


0.2
0.4
0.6
0.8
1.0

-1.0
-0.5
0.0
0.5
(a) (b)

Fig. C.1. (Constant injection rate, zero-viscosity) Zero- to the fth-order inertia terms of normalized opening, (a), and net-pressure, (b), in
small inertia expansion of the zero-viscosity solution.
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 479
I
a
(n; n
/
) =
4
p
_
n
/
G n; n
/
( )dn
/
=
4
p
2

1 n
2
_
sin
1
n
/
n
/
ln

1 n
/2
_

1 n
2
_

1 n
/2
_

1 n
2
_

_
n ln
n n
/
n n
/
1 nn
/

1 n
/2
_
1 n
2
_
1 nn
/

1 n
/2
_
1 n
2
_

_
; (D:4)
I
b
(n; n
/
) =
4
p
_
n
/
G(n; n
/
)n
/
dn
/
=
4
p

1 n
/2
_
1 n
2
_

n
/2
n
2
2
ln

1 n
/2
_

1 n
2
_

1 n
/2
_

1 n
2
_

_ _
. (D:5)
With the help of Eqs. (57) and (D.2), an expression for the function U = cos
1
n
4
p
_
1
n
dn in the lubrication
Eq. (53)
a
is given by
U(n; s) =cos
1
n
4
p
A(s)
A(0)
1
c
1
_
1
n
X
1
(n)dn

M1
i=1
[J
a
(1; n
/
) J
a
(n; n
/
)[a
i
(s) [J
b
(1; n
/
) J
b
(n; n
/
)[b
i
(s) [ [
n
/
=n
i1
n
/
=n
i
_ _
.
(D:6)
Lengthy closed-form expressions for the functions J
a,b
(n, n
/
) =
n
I
a,b
(n, n
/
)dn and
_
1
n
X
1
(n)dn are omitted here
for brevity.
Substitution of Eq. (D.6) into the lubrication Eq. (53)
a
and its evaluation at the mid-points n
i+1/2
=
(n
i
+ n
i+1
)/2 (i = 1, . . . , M 1) of the grid elements yields a system of M 1 coupled ordinary dierential
equations
4
p
_
A(s)
c
1
A(0)
_
1
n
X
1
(n)dn

M1
i=1
[J
a
(1; n
/
) J
a
(n; n
/
)[ _ a
i
(s) [J
b
(1; n
/
) J
b
(n; n
/
)[
_
b
i
(s)
_
n
/
=n
i1
n
/
=n
i
_ _
= 2(1 n
2
)
3=2
A(s)
A(0)
1
c
1
dP
1
dn
b
i
(s)
_ _
A(s)U(n; s); n = n
i1=2
; (D:7)
on M + 1 unknowns: A(s), dT
1
(s); . . . ; dT
M
(s). (Note that in Eq. (D.7), U is given by Eq. (D.6) and a
i
, b
i
are
dened in terms of dT
i
by Eq. (D.3).)
Two additional equations of the linear algebraic form are provided by the global continuity condition,
U
[n=0
= 0, Eq. (59), with (D.6) and (D.3), and the tip condition [Eq. (60)
b
]. The latter, given the properties
of the functions I
a,b
(n, n
/
) [Eqs. (D.4) and (D.5)], has the form

M1
i=1
[
^
I
a
(n
/
)a
i
(s)
^
I
b
(n
/
)b
i
(s)[
n
/
=n
i1
n
/
=n
i
=
p
2
5=2
A(s)
A(0)
1
_ _
; (D:8)
where
^
I
a;b
(n
/
) = lim
n1
(1 n)
1=2
I
a;b
(n; n
/
),
^
I
a
(n
/
) =
8

2
_
p
sin
1
n
/
;
^
I
b
(n
/
) =
8

2
_
p

1 n
/2
_
.
These two linear algebraic relations are used to reduce the number of unknowns in Eq. (D.7), i = 1, . . . , M 1,
by two (e.g., by expressing A(s) and dT
1
(s) in terms of dT
2
(s); . . . ; dT
M
(s)).
Initial conditions are provided by Eq. (58) in the form
dT
i
(0) = 0; i = 1; . . . ; M. (D:9)
Summarizing, the system of M 1 ODEs [Eq. (D.7)] plus two linear algebraic relations [Eqs. (59) and (D.8)],
and initial conditions [Eq. (D.9)] is solved numerically for A(s) and dT
1
(s); . . . ; dT
M
(s) by means of the ODE
solver in the software Mathematica, version 4.1 ( 19882000 Wolfram Research, Inc.).
The self-similar large-time asymptote of the above solution can be obtained either as the large-time limit
of the above numerical solution or can be solved directly. Indeed, this limit solution is time-independent (i.e.,
A and dT
1
; . . . ; dT
M
are asymptotic constants) and the right hand side of the lubrication Eq. (D.7) evaluated at
480 D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481
the grid element midpoint, n = n
i+1/2
, is zero. Thus, Eqs. (D.7), (59), and (D.8) yield a system of algebraic
equations on M + 1 unknown constants A and dT
1
; . . . ; dT
M
. This algebraic system is solved numerically
using Mathematica, which utilizes a damped Newtons method, the secant method and Brents method.
References
[1] Economides MJ, Nolte KG, editorsReservoir stimulation. 3rd ed. Chichester, UK: John Wiley & Sons; 2000.
[2] Spence DA, Turcotte DL. Magma-driven propagation of cracks. J Geophys Res 1985;90:57580.
[3] Detournay E, Garagash DI. The tip region of a uid-driven fracture in a permeable elastic solid. J Fluid Mech 2003;494:132.
[4] Nilson RH. Gas driven fracture propagation. ASME J Appl Mech 1981;48:75762.
[5] Nilson RH. Similarity solutions for wedge-shaped hydraulic fracture driven into a permeable medium by a constant inlet pressure. Int
J Numer Anal Methods Geomech 1988;12:47795.
[6] Spence DA, Sharp PW. Self-similar solution for elastohydrodynamic cavity ow. Proc Roy Soc London, Ser A 1985(400):289313.
[7] Lister JR. Buoyancy-driven uid fracture: The eects of material toughness and of low-viscosity precursors. J Fluid Mech
1990;210:26380.
[8] Carbonell RS, Desroches J, Detournay E. A comparison between a semi-analytical and a numerical solution of a two-dimensional
hydraulic fracture. Int J Solids Struct 1999;36(3132):486988.
[9] Garagash DI. Hydraulic fracture propagation in elastic rock with large toughness. In: Girard J, Liebman M, Breeds C,
Doe T, editors. Rock around the rimproceedings of the 4th North American rock mechanics symposium. Rotterdam: Balkema;
2000. p. 2218.
[10] Savitski AA, Detournay E. Propagation of a uid-driven penny-shaped fracture in an impermeable rock: Asymptotic solutions. Int J
Solids Struct 2002;39(26):631137.
[11] Garagash DI, Detournay E. Plane-strain propagation of a uid-driven fracture: small toughness solution. ASME J Appl Mech
2005;72:91628.
[12] Detournay E. Fluid and solid singularities at the tip of a uid-driven fracture. In: Detournay D, Pearson JRA, editors.
Proceedings of the IUTAM symposium on non-linear singularities in deformation and ow, Haifa. Dordrecht: Kluwer Academic
Publ.; 1999. p. 2742.
[13] Garagash DI, Detournay E. Viscosity-dominated regime of a uid-driven fracture in an elastic medium. In: Karihaloo BL, editor.
IUTAM symposium on analytical and computational fracture mechanics of non-homogeneous materials, Cardi. Solid mechanics
and its applications, vol. 97, March 2002. p. 259.
[14] Adachi JI, Detournay E. Self-similar solution of a plane-strain fracture driven by a power-law uid. Int J Numer Anal Methods
Geomech 2002;26:579604.
[15] Adachi JI. Fluid-driven fracture in permeable rock. PhD thesis, University of Minnesota, 2001.
[16] Bunger AP, Detournay E, Jerey RG. Crack tip behavior in near-surface uid-driven fracture experiments. CR Acad Sci Paris
2005;333:299304.
[17] Huang N, Szewczyk A, Li Y. Self-similar solution in problems of hydraulic fracturing. ASME J Appl Mech 1990;57:87781.
[18] Batchelor GK. An introduction to uid dynamics. Cambridge, UK: Cambridge University Press; 1967.
[19] Rice JR. Mathematical analysis in the mechanics of fracture. In: Liebowitz H, editor. Fracture, an advanced treatise, vol. II. New
York, NY: Academic Press; 1968. p. 191311 [Chapter 3].
[20] Papanastasiou P. The eective fracture toughness in hydraulic fracturing. Int J Fracture 1999;96:12747.
[21] Shapiro RA. The dynamics and thermodynamics of compressible uid ow, vol. II. New York: Ronald Press; 1954.
[22] Sneddon IN, Lowengrub M. Crack problems in the classical theory of elasticity. New York, NY: John Wiley & Sons; 1969.
[23] Nilson RH, Morrison FA. Transient gas or liquid ow along a preexisting or hydraulically-induced fracture in permeable medium.
ASME J Appl Mech 1986;53:15765.
[24] Detournay E. Propagation regimes of uid-driven fractures in impermeable rocks. Int J Geomech 2004;4(1):111.
[25] Van Dyke M. Perturbation methods in uid mechanics. Stanford: Parabolic Press; 1975.
[26] Liskovets OA. The method of lines. J Dier Eqs 1965;1:1308.
[27] Nilson RH, Griths SK. Numerical analysis of hydraulically-driven fractures. Comput Methods Appl Mech Eng 1983;36:35970.
[28] Garagash DI, Detournay E. An analysis of the inuence of the pressurization rate on the borehole breakdown pressure. Int J Solids
Struct 1997;34(24):3099118.
D.I. Garagash / Engineering Fracture Mechanics 73 (2006) 456481 481

You might also like