You are on page 1of 9

http://www.paper.edu.

cn

Computers and Geotechnics 32 (2005) 2735


www.elsevier.com/locate/compgeo

Application of modied EP method in steady seepage analysis


Chun-Ning Ji a, Yuan-Zhan Wang

a,*

, Ying Shi

School of Civil Engineering, Tianjin University, Tianjin 300072, PR China


Huahui Architectural Design & Engineering Co. Ltd., Tianjin 300381, PR China

Received 22 April 2004; received in revised form 28 October 2004; accepted 22 November 2004
Available online 18 January 2005

Abstract
Free-surface (unconned) seepage problems are commonly encountered in geotechnical engineering. In these problems, the determination of the free-surface usually requires sophisticated numerical techniques. Among all the proposed methods for solving freesurface seepage problems, the extended pressure (EP) method emerges as one of the easiest methods since the boundary conditions
are simply imposed on the prescribed outer boundaries and those on the prior unknown free-surface are automatically satised.
However, in the available papers on the EP method, the optimal value of e in the ramp function He(p), which has signicant eects
on solving accuracy and computational eciency, is rarely mentioned or simply selected as the grid spacing without a systematical
analysis. In this paper, a modied EP method is proposed to improve the accuracy in simulating the free-surface and reduce the
computational cost. In this modied EP method, e is proposed as half of the under-grid spacing in y-direction through an error
analysis of the nite dierence equations and their iteration schemes. Compared with the original EP method, the modied EP
method is more accurate and ecient due to its smooth free-surface, rational spillpoint position and low computational cost.
2005 Elsevier Ltd. All rights reserved.
Keywords: Seepage; Free-surface; Modied extended pressure method; Finite dierence

1. Introduction
Free-surface (unconned) seepage problems are
commonly encountered in the practice of geotechnical
engineering and many robust numerical methods have
been proposed to solve these problems. However,
these numerical methods are not commonly used in
engineering practice and largely ignored in soil
mechanics textbooks, mainly because they require
rather complicated derivations and implementations.
Some textbooks (e.g. [1]) present basic spreadsheet
solutions for conned seepage problems with multiple
soil layers and anisotropic permeability. However,
these basic spreadsheet techniques are limited to con-

Corresponding author. Tel./fax: +86 22 27401337.


E-mail address: yzwang@public.tpt.tj.cn (Y.-Z. Wang).

0266-352X/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compgeo.2004.11.006

ned seepage problems for which the ow boundaries


are known. At the present, they do not apply to
unconned seepage problems, such as free-surface
ows within earthdams. Therefore, there is a requirement to propose a simple and ecient numerical
method for practical engineering and educational
training.
In free-surface seepage problems, the key point is
the determination of the free-surface which delimits
the ow boundaries and can be found using nonlinear numerical techniques including nite dierence
method with adaptive mesh [2], nite element method
with adaptive mesh (e.g. [35]) and xed mesh (e.g.
[614]). Among all proposed methods, the extended
pressure (EP) method, initially proposed by Brezis
et al. [15] and further solved by Jean et al. [16] using
nite dierence method, is one of the simplest and
most ecient methods. Through an extension of


28

http://www.paper.edu.cn
C.-N. Ji et al. / Computers and Geotechnics 32 (2005) 2735

Darcys law, the EP method reduces variational


inequalities to simpler equalities which are then applied in the whole computational domain. This suggests that the boundary conditions can be simply
imposed on the prescribed outer boundaries without
sophisticated techniques, and those on the prior unknown free-surface are automatically satised in the
solving process. As indicated in Brezis et al. [15],
the parameter e in the ramp function He(p) has signicant eects on solving accuracy and computational
eciency. However, in all available related papers,
the optimal value of e is rarely mentioned or simply
selected as the grid spacing (Jean et al. [16]) without
a systematical analysis. In order to improve the solving accuracy and computational eciency of unconned seepage problems using the EP method, a
reasonable selection of e is proposed by an error
analysis of the nite dierence equations and their
iteration schemes.

2. Theory
2.1. Review of seepage theory
Two-dimensional ow of water through porous soils
is assumed to follow Darcys law [17]:
v K  gradh;

where v is the velocity vector, K is the permeability matrix, h is the total head.
The conservation of water mass in the control volume
X leads to the following governing equation:
Z
v  n ds 0;
2
S

where S is the boundary of X and n is the unit vector


normal to S. Eqs. (1) and (2) lead to the well-known partial dierential equation
kx

o2 h
o2 h

k
0;
y
ox2
oy 2

which further reduces to Laplace equation in the case of


isotropic permeability (kx = ky).
2.2. Finite dierence equations for conned seepage
The nite dierence equations for seepage can be formulated (1) from Eq. (3) using the second-order derivatives of h, or (2) more eciently from Eq. (2) using the
rst-order derivatives of h. As shown in Fig. 1, when S
is taken as A2A4A6A8 for irregular and rectangular nite
dierence grids, Eq. (2) leads to the following nite difference equation:

Fig. 1. Grid geometry used for derivation of nite dierence


equations.

hi;j  hi1;j
hi1;j  hi;j
k x1 S 1 k x8 S 8 
k x4 S 4 k x5 S 5
Dx1
Dx2
hi;j  hi;j1
hi;j1  hi;j

k y2 S 2 k y3 S 3 
k y6 S 6 k y7 S 7 0;
Dy 1
Dy 2
4

where hi,j, hi  1,j, hi + 1,j, hi,j  1 and hi,j + 1 are the values
of total head h at grid nodes; S1, S2, . . . ,S8 are the segments A1A2, A2 A3, . . . ,A8A1; Dx1 and Dx2 are the leftand right- grid spacing in the x-direction, respectively;
Dy1 and Dy2 are the up- and down-grid spacing in the
y-direction, respectively; kx1, kx4, kx5 and kx8 are the
permeability in the x-direction in the four quadrants
surrounding the center node and ky2, ky3, ky6 and ky7
are the corresponding permeability in the y-direction.
Eq. (4) can be rewritten as follows:

1
k x1 S 1 k x8 S 8
k x4 S 4 k x5 S 5
hi1;j
hi1;j
hi;j
D
Dx1
Dx2

k y2 S 2 k y3 S 3
k y6 S 6 k y7 S 7
;
5
hi;j1
hi;j1
Dy 1
Dy 2
where the coecient D is
D

k x1 S 1 k x8 S 8 k x4 S 4 k x5 S 5 k y2 S 2 k y3 S 3

Dx1
Dx2
Dy 1
k y6 S 6 k y7 S 7

:
Dy 2

Eq. (5) is second-order accurate when the element


A2A4A6A8 is selected such as S2 = S7 = Dx1/2;
S3 = S6 = Dx2/2; S1 = S4 = Dy1/2 and S5 = S8 = Dy2/2
because the rst-order derivatives of h are second-order
accurate when evaluated at the middle of grid points.
Through the selection of coecients Si, kxi, kyi, Dxi
and Dyi, Eq. (5) encompasses many cases of anisotropic

http://www.paper.edu.cn
C.-N. Ji et al. / Computers and Geotechnics 32 (2005) 2735

permeability, layered permeability and impervious


boundary conditions.
2.3. EP method governing equation and its nite dierence
equations for unconned seepage
Free-surface seepage problems can be dened
through the generic problem in Fig. 2. The saturated domain Xw is comprised by points ABCDE whereas the
dry domain Xd is located above the free-surface AE,
which has a prior unknown location. The governing
equation and boundary conditions of the free-surface
seepage problems of Fig. 2 are:
divv 0; p P 0; in Xw ;

v K  gradp y;

p 0; in Xd ;

p hAB  y; on AB and p hCD  y; on CD;

10

n  v 0; on BC;

11

n  v 0; p 0; on AE;

12

p 0; on DE;

13

where p is the pressure head; y is the grid altitude; n represents the unit vector normal to the boundary; hAB and
hCD are the water depth on the left and right side, respectively. In the solution of free-surface seepage problems, it
is convenient to take the pressure head p as unknown instead of the total head h. The pressure head p can be described as p = h  y and related to water pressure u and
water unit weight cw through p = u/cw.
In the concept of extended pressure, Brezis et al. [15]
modied Darcys relation as follows:
v0 K  gradp H pgrady;

14

where v 0 is the modied seepage velocity, H(p) is the


Heaviside function:

1; if p P 0;
H p
15
0; if p < 0:

29

When p is positive, Eq. (14) becomes identical to Eq.


(8). In the EP framework, the new governing equation
and boundary conditions for seepage problems with a
free-surface are made of Eqs. (10), (11), (14), (16)
and (17).
divv0 0; in Xw [ Xd ;

16

p 0; on DEFA:

17

In Eq. (14), the modied velocity composes of two


dierent parts, one part introduced by pressure and
the other introduced by gravity. Because of the absence of water, the gravity has no eect on the velocity
in the dry domain where H(p) = 0. But in the wet domain where H(p) = 1, the contribution of gravity
should be taken into account to balance the impact
of pressures gradient in y-direction. Thus, the Heaviside function H(p) in Eq. (14) just likes a switch to
reect the change of material attributes in the whole
computational domain. In the concept of EP method,
the free-surface is considered as a material interface
and the boundary conditions on it are automatically
satised by adjusting the Heaviside function H(p).
Furthermore, the mathematical derivations that establish the equivalence between the EP boundary value
problem and the original free-surface seepage problem
can be found in Brezis et al. [15] and Oden and Kikuchi [10].
Since the value of the Heaviside function H(p) varies
from 0 to 1 abruptly at two sides of the free-surface,
numerical oscillations may appear in solving process.
For this reason, it is convenient to replace the Heaviside
function H(p) with the following ramp function He(p):

1; if p P e;
18
H e p
p=e; if p < e:
When e tends towards zero, He(p) tends towards H(p).
The nite dierence equations of EP method for
unconned seepage, like those for conned seepage,
can be conveniently derived using an area S surrounding
a grid node:
Z
Z
Z
K  gradp ds k y H e pny ds 0:
 v0  n ds
S

19
For the grid shown in Fig. 1, the last integral of Eq. (19)
is evaluated as follows:


Z
pi;j pi;j1
k y H e pny ds k y2 S 2 k y3 S 3 H e
2
S


pi;j pi;j1
:
 k y6 S 6 k y7 S 7 H e
2
20
Fig. 2. A generic seepage problem with a free surface.

Therefore, the nite dierence equation for unconned


seepage is


30

pi;j

http://www.paper.edu.cn
C.-N. Ji et al. / Computers and Geotechnics 32 (2005) 2735


1
k x1 S 1 k x8 S 8
k x4 S 4 k x5 S 5
pi1;j
p
Dx1
Dx2
D i1;j

k y2 S 2 k y3 S 3
k y6 S 6 k y7 S 7
pi;j1
pi;j1
Dy 1
Dy 2



pi;j pi;j1
1
k y2 S 2 k y3 S 3 H e

2
D


pi;j pi;j1
:
 k y6 S 6 k y7 S 7 H e
2

21

Compared to Eq. (5) for conned seepage, Eq. (21) for


unconned seepage has nonlinear He terms, which render it slightly more complicated. Eq. (21) becomes linear
and coincides with Eq. (5) when the values of
(pi,j  1 + pi,j)/2 and (pi,j + pi,j + 1)/2 are larger than e.
2.4. SOR solution
To accelerate the convergence process, the successive
over-relaxation (SOR) is applied in solving Eqs. (5) and
(21). For instance, the SOR version of Eq. (5) is

x m k x1 S 1 k x8 S 8
m

1

xh

h
hm1
i;j
i;j
Dx1
D i1;j
k
S

k
S
k y2 S 2 k y3 S 3
x4 4
x5 5
hmi;j1
hmi1;j
Dx2
Dy 1

k
S

k
S
y6 6
y7 7
;
22
hmi;j1
Dy 2
where x is the relaxation factor, the value of which is
usually taken between 1 and 2 for faster convergence;
hmi;j represents the value of hi,j at the mth iteration. In
the iteration process, the index i is increasing whereas
is calcuthe index j is decreasing. Therefore, when hm1
i;j
m1
and
h
are
known.
So,
the
nal
nite
diflated, hm1
i1;j
i;j1
ference equation for conned seepage problems is

x m1 k x1 S 1 k x8 S 8
m
h

1

xh

hm1
i;j
i;j
Dx1
D i1;j
k
S

k
S
k y2 S 2 k y3 S 3
x4 4
x5 5
hmi;j1
hmi1;j
Dx2
Dy 1

k
S

k
S
y6 6
y7 7
:
23
hm1
i;j1
Dy 2
Similarly, the SOR version of Eq. (21) is

x m k x1 S 1 k x8 S 8
m
p

1

xp

pm1
i;j
i;j
Dx1
D i1;j
k
S

k
S
k y2 S 2 k y3 S 3
x4 4
x5 5
pmi;j1
pmi1;j
Dx2
Dy 1

k
S

k
S
y6 6
y7 7
pmi;j1
Dy 2

 m

pi;j pmi;j1
x

k y2 S 2 k y3 S 3 H e
2
D
 m

pi;j pmi;j1
;
 k y6 S 6 k y7 S 7 H e
2

24

where pmi;j represents the value of pi,j at the m th iteration.


The iteration process of Eq. (24) is same as that of Eq.
(22). The nal nite dierence equation for the unconned seepage problems is

x m1 k x1 S 1 k x8 S 8
m1
m
p
pi;j 1  xpi;j
Dx1
D i1;j
k x4 S 4 k x5 S 5
k y2 S 2 k y3 S 3
pmi;j1
pmi1;j
Dx2
Dy 1

k y6 S 6 k y7 S 7
pm1
i;j1
Dy 2

 m

pi;j pmi;j1
x

k y2 S 2 k y3 S 3 H e
2
D
!!
m1
m
pi;j pi;j1
 k y6 S 6 k y7 S 7 H e
:
25
2
2.5. Eects of e on solving accuracy and computational
eciency
Brezis et al. [15] indicated that the parameter e in their
proposed the ramp function He(p) has signicant eect
on the solution of EP method governing equation. Also,
Oden and Kikuchi [10] remarked that the EP method
converges when e is selected not smaller than the grid
spacing in y-direction. Jean et al. [16] obtained a numerical solution by using the ramp function He(p) proposed
by Brezis, assuming e as the grid spacing in the y-direction. But all above conclusions have not been veried by
a systematic analysis yet.
To inspect the eects of the parameter e on solving
accuracy and computational eciency, a one-dimensional and a two-dimensional steady isotropic seepage
problems are solved using EP method.

Fig. 3. One-dimensional water column with an unknown free surface:


(a) geometry and boundary conditions; (b) comparison of solutions
obtained by the EP method with dierent e.

http://www.paper.edu.cn
C.-N. Ji et al. / Computers and Geotechnics 32 (2005) 2735

In the one-dimensional seepage problem, the mesh is


regular, assuming uniform grid spacing as 0.5 m, top
boundary condition as p = 0.0 m and bottom boundary
condition as p = 5.0 m, as shown in Fig. 3(a). The
parameter e is chosen as 0.25, 0.50, 1.00 and 2.00 m,
respectively. The relaxation factor x is 1.8 and the permissible iteration tolerance is 0.0001. The numerical
solutions with dierent e are shown in Fig. 3(b).
It is obvious that the physical solution for the above
one-dimensional isotropic steady seepage problem is:
p = 0, when y < 5 m ; p = y  5, when y P 5 m. As
shown in Fig. 3(b), when e = 0.25 m (i.e., e = Dy2/2),
the pressure head curve is discontinuous at the free-surface, which conforms to the physical solution. With
increasing value of e, the numerical solution displays a
smooth transition of pressure head gradient between
wet and dry domain, instead of a discontinuous slope.
The EP method therefore does not dene exactly the position of the free-surface, when e is larger than 0.25 m.
The convergence process of the above one-dimensional seepage problem is illustrated in Fig. 4, assuming
e = 0.25 m. Apparently, the numerical solution converges towards the physical solution rapidly. After 20
iterations, the error between physical and numerical
solutions on the free-surface becomes smaller than
0.0003. The numbers of iterations required to achieve

31

Table 1
Values of parameters used in numerical examples: grid spacing, EP
parameter e, SOR parameter x, number of iterations required to
achieve convergence (tolerance is 0.0001 in all cases), error between
physical and numerical solutions on free-surface (only for onedimensional water column)
Figure no.

Grid
spacing

Number of
iterations

Error
(smaller than)

0.5

0.25
0.5
1.0
2.0

1.8

20
25
38
49

0.0003
0.1666
0.3554
0.6629

0.5

0.25
0.5
1.0

1.7

27
32
73

N.A.

6
7

0.25
0.25

0.125
0.125

1.7
1.5

39
102

N.A.
N.A.

Note: x is the optimal value from 1 to 2.

convergence with dierent e are listed in Table 1. With


increasing value of e, more iteration steps should be executed to obtain the convergence of solution.
In the two-dimensional seepage problem, the mesh
is regular with even grid spacing assumed as 0.5 m
in the x- and y-directions and the boundary conditions
are: left water depth is h = 10.0 m, right water depth is
h = 2.0 m and bottom boundary is impermeable, as
shown in Fig. 5(a). The parameter e is chosen as
0.25, 0.50 and 1.00 m, respectively. The relaxation factor x is 1.7 and the permissible iteration tolerance is
0.0001. To validate the solutions of EP method, a
solution of the Geo-slope with same geometry and
boundary conditions, assumed as the physical solution,
is also presented.
As shown in Fig. 5(b), the location of the free-surface
computed by the proposed method is very similar to that
of Geo-slope, when e = 0.25 m (i.e., e = Dy2/2). Whereas,
while the parameter e increases, the error between the
solutions of the EP method and Geo-slope increases
rapidly and the spillpoints elevation obtained by the
EP method becomes much smaller than that of Geoslope. Otherwise, as shown in Table 1, when e equals
to 0.25 m, 27 iteration steps should be executed before
solving process converges. When e is larger than 0.25
m, much more eorts should be performed to achieve
convergence.
From above, it is evident that the parameter e has signicant eects on solving accuracy and computational
eciency. An optimal value of e can not only improve
the solving accuracy but also reduce the computational
cost of the EP method.
2.6. Selection of e in ramp function He(p)

Fig. 4. Variation of computed pressure head p distribution with


iteration number and physical solution in the case of the onedimensional water column example.

To determine the optimal value of e, the mechanism


of its eect on solving accuracy is analyzed.


32

http://www.paper.edu.cn
C.-N. Ji et al. / Computers and Geotechnics 32 (2005) 2735

Fig. 5. Two-dimensional rectangle dam with an unknown free surface: (a) geometry and boundary conditions; (b) comparison of solutions obtained
by the EP method with dierent e and Geo-slope.

For one-dimensional isotropic steady seepage problems, Eq. (21) can be rewritten as follows:
1
pi pi1 pi1
2
p p 
p p 
1
i1
i1
Dy 1 H e i
 Dy 2 H e i
:
2
2
2

26

Assuming the position of the free-surface is at y = yi, the


physical solution is: pi  1 = pi = 0, pi + 1 = Dy2. Then,
Eq. (26) further reduces to



Dy 2 1
Dy 2

:
27
Dy 1 H e 0  Dy 2 H e
pi
2
2
2
When e = Dy2/2, Eq. (27) becomes pi = 0, which conforms to the physical solution. Whereas, when e > Dy2/
2, it becomes pi Dy 2 =21  Dy 2 =2e > 0, which is not
coincide with the actual case. Moreover, when e < Dy2/
2, the solving process will lead to divergence.
For two-dimensional isotropic steady seepage problems, Eq. (21) can be rewritten as follows:



pi1;j pi1;j
k
Dy 1 Dy 2

pi;j
Dx1
Dx1
2D


pi;j1 pi;j1

Dx1 Dx2
Dy
Dy 2
 1

pi;j pi;j1
Dx1 Dx2 H e
2


pi;j pi;j1
:
28
 He
2

Supposing the grid point Pi,j is on the free-surface, the


physical solution is: pi,j  1 = pi,j = 0, pi,j + 1 = Dy2. As
the grid points Pi  1,j, Pi + 1,j and Pi,j are on the same
altitude level, the pressure head on the three grid points
are approximately equal (i.e., pi  1,j = pi + 1,j = pi,j = 0),
when the grid spacing in x-direction and the gradient
of free-surface are not very large. Then, Eq. (28) can
be further reduces to



kDx1 Dx2
Dy 2
:
29
1 H e 0  H e
pi;j
2
2D
When e = Dy2/2, Eq. (29) becomes pi,j = 0, which is
accordant with the physical solution. Whereas, when
e > Dy2/2, it becomes


kDx1 Dx2
Dy
1  2 > 0;
pi;j
2e
2D
which does not conform to the actual case. Moreover,
when e < Dy2/2, the solving process will lead to
divergence.
As to anisotropic steady seepage problems, the same
conclusion can also be obtained through similar
analysis.
From above, it can be concluded that the numerical
solution obtained by the EP method is coincident with
physical solution, when the parameter e is selected as
half of the under-grid spacing in y-direction (i.e.,
e = Dy2/2).

http://www.paper.edu.cn
C.-N. Ji et al. / Computers and Geotechnics 32 (2005) 2735

3. Applications
For the complexity of geometry and the nonhomogeneous permeability in the following two steady seepage problems, no analytical solution is presented in the
available papers. So, the solutions of Lacy and Prevost
[18], whose accuracy has been veried by practices, are
considered as the physical solutions.
3.1. Isotropic dam with a slanted face
Fig. 6 shows the geometry of a dam with a slanted
downstream surface. Also, the comparison of the freesurface obtained by proposed method and other methods is illustrated. The values used in numerical solving
are listed in Table 1. From the numerical solutions in
Fig. 6, the location of free-surface computed by the proposed method is very similar to that presented by Lacy
and Prevost [18]. Compared with the solution of the original EP method (Jean and Tetsuo [16]), the solving

33

accuracy on the free-surface is substantially improved


for rational spillpoint position. Otherwise, as shown in
Table 2, the iteration number to achieve convergence
of the modied EP method is much smaller than that
of original EP method.
3.2. Nonhomogeneous rectangular dam
Fig. 7 shows the geometry of a dam made of two rectangular blocks of dierent permeability. The downstream block is ten times more pervious than the
upstream one (i.e., k2/k1 = 10). For this reason, numerical oscillations may appear on the material interface,
which is the major cause of the deterioration of solving
accuracy and computational eciency. The values used
in numerical solving are listed in Table 1. As shown in
Fig. 7, the location of free-surface computed by the
proposed method agrees well with that reported by
Lacy and Prevost [18]. Compared with the solution of
the original EP method (Jean and Tetsuo [16]), the

Fig. 6. Comparison of free-surface of isotropic dam with a slanted face obtained by proposed method, Jean and Tetsuo [16], Borja and Kishnani [19]
and Lacy and Prevost [18].

Table 2
Comparison of model parameters and number of iterations for original and modied EP methods
Figure no.

Method

Grid spacing

Jean and Tetsuo [15]


Present method

0.5

0.5
0.25

1.8
1.8

25
20

Jean and Tetsuo [15]


Present method

0.5

0.5
0.25

1.2
1.7

135
27

Jean and Tetsuo [15]


Present method

0.25

0.25
0.125

1.2
1.7

325
39

Jean and Tetsuo [15]


Present method

0.25

0.25
0.125

1.2
1.5

414
102

Note: x is the optimal value from 1 to 2.

Number of iterations


34

http://www.paper.edu.cn
C.-N. Ji et al. / Computers and Geotechnics 32 (2005) 2735

the EP method is that nite dierence grids can only


be rectangular or triangular, therefore excluding curved
boundaries.

5. Conclusion

Fig. 7. Comparison of free-surface of nonhomogeneous rectangular


dam obtained by proposed method, Jean and Tetsuo [16], Borja and
Kishnani [19] and Lacy and Prevost [18].

computational eciency of the modied EP method is


improved substantially for the less iterations required
to achieve convergence listed in Table 2.
From above, it can be concluded that, compared with
the solutions obtained by the original EP method, the
solving accuracy and computational eciency of the
modied EP method are substantially improved for
smooth free-surface, reasonable spillpoint position and
low computational cost.

A nite dierence approach has been proposed for


calculating the solutions of unconned seepage with an
unknown free-surface. The proposed method is based
on the EP method which modies the Darcys relation
and applies the governing equation in the whole computational domain. In the EP framework, the free-surface
boundary conditions are automatically satised, which
simplies the solving process. The error analysis shows
that the parameter e in the ramp function He(p) has signicant eects on solving accuracy and computational
eciency. By analyzing the nite dierence equations
and their iteration schemes, the optimal value for e is
proposed as half of the under-grid spacing in y-direction
(i.e., e = Dy2/2). Compared with the results of the original EP method reported by Jean et al., the solving accuracy and computational eciency of the modied EP
method are substantially improved due to smooth freesurface, rational spillpoint position and low computational cost. The proposed method has not only
educational values as it describes openly the equations
and processes for solving free-surface seepage problems,
but also practical values as it can solve many seepage
problems with high solving accuracy and low computational cost.

Acknowledgement
The work presented herein has been supported by the
National Natural Science Foundation of China under
Grant No. 50279027.

References
4. Discussion
The main advantage of the EP method is its ease of
implementation and application to a wide variety of
practical free-surface seepage problems. From an educational point of view, all the assumptions, equations and
calculation steps are clearly stated, formulated and executed, which is rather uncommon for other non-linear
numerical techniques for free-surface seepage problems.
The proposed method is not claimed to compete with
more versatile commercial seepage programs. However,
all the calculation examples in Figs. 37 were completed
within a few minutes on personal computers, making it
suitable for engineering practice. Another limitation of

[1] Bardet JP. Experimental soil mechanics. Upper Saddle River,


NJ: Prentice-Hall; 1997.
[2] Cryer CW. On the approximate solution of free boundary
problems using nite dierences. J Assoc Comput Mach
1970;17(3):397411.
[3] Taylor RL, Brown CB. Darcy ow solutions with a free surface. J
Hydr Div, ASCE 1967;93:2533.
[4] Finn WDL. Finite-element analysis of seepage through dams. J
Soil Mech Foundat Div, ASCE 1967;93(SM6):418.
[5] Neuman SP, Witherspoon PA. Finite element method of analyzing steady seepage with a free surface. Water Resour Res
1970;6(3):88997.
[6] Baiocchi C. Su un problema di frontiera libera connesso a
questioni di idraulica. Ann Math Pura Appl 1972;91:10727.
[7] Bathe KJ, Khoshgoftaar MR. Finite element free surface seepage
analysis without mesh iteration. Int J Numer Anal Meth
Geomech 1979;3:1322.

http://www.paper.edu.cn
C.-N. Ji et al. / Computers and Geotechnics 32 (2005) 2735

[8] Kikuchi N. An analysis of the variational inequalities of seepage


ow by nite-element methods. Quart Appl Math
1977;35:14963.
[9] Alt HW. Numerical solution of steady-state porous ow free
boundary problems. Numer Math 1980;31:7398.
[10] Oden JT, Kikuchi N. Recent advances: theory of variational
inequalities with applications to problems of ow through porous
media. Int J Eng Sci 1980;18:1173284.
[11] Friedman A. Variational principles and free-boundary problems. New York: Wiley; 1982.
[12] Desai CS, Li GC. A residual ow procedure and application for
free surface ow in porous media. Adv Water Resour
1983;6:2735.
[13] Baiocchi C, Capelo A. Variational and quasivariational inequalities. Applications to free boundary problems. New York: Wiley;
1984.

35

[14] Westbrook DR. Analysis of inequalities and residual ow


procedures and an iterative scheme for free surface seepage. Int
J Numer Meth Eng 1985;21:1791802.
[15] Brezis H, Kinderlehrer D, Stampacchia G. Sur une nouvelle
formulation du probleme de lecoulement a travers une digue,
Serie A. Paris: C. R. Academie des Sciences; 1978.
[16] Jean-Pierre Bardet, Tetsuo Tobita. A practical method for solving
free-surface seepage problems. Comput Geotech 2002;29:45175.
[17] Darcy H. Les Fontaines publiques de la ville de
Dijon. Paris: Dalmont; 1856.
[18] Lacy SJ, Prevost JH. Flow through porous media: a procedure for
locating the free surface. Int J Numer Anal Meth Geomech
1987;11:585601.
[19] Borja RI, Kishnani SS. On the solution of elliptic free-boundary
problems via Newtons method. Comput Meth Appl Mech Eng
1991;88:34161.

You might also like