Analytical solutions for seepage near material boundaries in dam cores: The

DavisonKalinin problems revisited

Anvar Kacimov
a,
, Yurii Obnosov
b
a
Department of Soils, Water and Agricultural Engineering, Sultan Qaboos University, Oman
b
Institute of Mathematics and Mechanics, Kazan University, Russia
a r t i c l e i n f o
Article history:
Received 3 September 2010
Received in revised form 14 July 2011
Accepted 27 July 2011
Available online 7 September 2011
Keywords:
Analytic functions
Free boundary problems
Lapalces equation
Seepage
Refraction
Hydraulic gradient
a b s t r a c t
Steady Darcian seepage through a dam core and adjacent shells is analytically studied. By
conformal mappings of the pentagon in the hodograph plane and triangle in the physical
plane ow through a low-permeable dam core is analyzed. Mass-balance conjugation of
ow in the core and downstream highly-permeable shell of the embankment is carried
out by matching the seepage ow rates in the two zones assuming that all water is inter-
cepted by a toe-drain. Seepage refraction is studied for a wedge-shaped domain where
pressure and normal components of the Darcian velocities coincide on the interface
between the core and shell. Mathematically, the problem of R-linear conjugation (the
RiemannHilbert problem) is solved in an explicit form. As an illustration, ow to a
semi-circular drain (lter) centered at the triple point (contact between the core, shell
and impermeable base) is studied. A piece-wise constant hydraulic gradient in two adja-
cent angles making a two-layered wedge (the dam base at innity) is examined. Essentially
2-D seepage in a domain bounded by an inlet constant head segment, an outlet seepage-
face curve, a horizontal base and with a straight tilted interface between two zones (core
and shell) is investigated. The ow net, isobars, and isotachs in the core and shell are recon-
structed by computer algebra routines as functions of hydraulic conductivities of two
media, the angle of tilt and the hydraulic head value at a specied point.
2011 Elsevier Inc. All rights reserved.
1. Introduction
Renewed interest to hydropower stations, dam reservoirs and large water supply schemes drives both civil/geotechncial
engineers and applied mathematicians, dealing with movement of water through porous materials, to revisit the legacy of
the founders of the specialism of subsurface mechanics: Bear, Casagrande, Cedergren, Dachler, Davison, Charny, Gersevanov,
Hamel, Muscat, Numerov, Pavlovsky, Riesenkampf, Polubarinova-Kochina [1]. What unies these towering gures? All of
them 50100 years ago worked on mathematical problems of seepage through earth dams and their contribution is now
in both manuals of geotechnical engineers and applied math books, where the damproblem became a shibboleth of a com-
munity of mathematicians working with free boundary problems (e.g., [2]). The objective of this paper is to amend solutions
to the Laplace equation for the dam problem by an analytical study of seepage through dam heterogeneities.
Barrages, dikes, levees, weirs and embankments demarcating reservoirs, ponds, detention pools, canals and other hydrau-
lic and agro-engineering structures involve often an earth-, rock-lled element, which maintains a difference of water levels
on two sides of the structure. Dams made of local porous materials are cheap but permeable to water that seeps through the
0307-904X/$ - see front matter 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2011.07.088

Corresponding author. Fax: +968 24413 418.

E-mail addresses: anvar@squ.edu.om (A. Kacimov), yobnosov@ksu.ru (Y. Obnosov).
Applied Mathematical Modelling 36 (2012) 12861301
Contents lists available at SciVerse ScienceDirect
Applied Mathematical Modelling
j our nal homepage: www. el sevi er . com/ l ocat e/ apm
dam soil from the upper pool (reservoir) with a higher water level to the tail water with a lower level. Permeable dams
according to the International Commission on Large Dams inventory make more than 80% of newly constructed large
structures in the world and the vast majority of smaller impoundments. In Oman, after the June-2007 Gonu cyclone and dev-
astating oods
1
the government immediately invested into building new and upgrading the existing (31 as of 2009) protection,
groundwater recharge and storage dams, in particular, the Wadi Adei cascade of eight dams, Al-Khod dam and the rst large
(70 m high, 100 mln m
3
of reservoir capacity) Wadi Dayqah dam.
The tailing-dams have the length of up to several kilometers (e.g. the Al-Khod dam is 5.1 km long) and commonly consist
of the so-called side shells (shoulders), which offer structural resistance against failure, but have a relatively high
hydraulic conductivity k
1
(i.e. little hydraulic resistance against seepage) and a relatively expensive core composed of a
material of low conductivity k
2
, which serves as a seepage-checking element (barrier, see, e.g. [3]). A typical vertical
cross-section is shown in Fig. 1 where the upper pool water level H
1
is translated almost without any loss through the up-
stream shell to the left face AB of the core. From the right face DC of the core again almost no loss of the head occurs through
a coarse lling of the downstream shell to the tailwater, where the water level is H
2
.
Both the downstream and upstream shoulders may consist of two or more zones (e.g. an inner shell and outer shell) of
contrasting conductivity and then seepage occurs through a composite medium of conductivities k
1
, k
2
, k
3
, . . . Seepage in
downstream shell is often intercepted by toe drains such that a phreatic surface DC
1
does not emerge on the slope of the
tailwater (Fig. 1). Water seeped through an element of a dam and collected by a drain is usually diverted to gutters installed
in a dam gallery. From there the leachate is discharged either gravitationally or by pumping to the tailwater.
The right face of the core (ADC in Fig. 1) is commonly subject to seepage erosion. A chimney drain (Fig. 1) or diaphragm is
then installed [4,5], forestalling a direct contact between the clay core and coarse shell and dislodging/migration of ne par-
ticulate into the downstream shell. The lack or fouling/degradation of lters may trigger sand boiling (as, e.g., with the New
Orleans levees) that may lead to a collapse with multi-billion damage and require potentially expensive repercussions.
Dam lters are often graded according to the Vicksburg Lab method [6] or an earlier published Soviet instructions for de-
sign of lters (see [7] for review), which stipulate that d
f
60
=d
f
10
6 6, and d
f
10
=d
s
60
6 6 where d
f
and d
s
are the particle diameters
retrieved from the particle-size distribution curves of the lter and suffosion-prone dam material, correspondingly. The Ter-
zaghi-layering and gradation of the lter material according to US Department of Agriculture/American Society for Testing
Notations and nomenclature
b, b
d
the widths of the downstream shell
G
z
, S
1
, S
2
, G
w
, G
V
, G
f
domains and subdomains in the physical, complex potential, hodograph and auxiliary planes (corre-
spondingly)
h hydraulic head
H head drop between the upper pool and tailwater
H
1
, H
2
water levels in the upper pool and tailwater
k
1
, k
2
, k
f
hydraulic conductivities of conjugated dam zones
k
r
, K
r
conductivity ratios
L, L
1
the core and shell widths
p pressure head
Q seepage ow rate through the dam (per unit width)
v
x
, v
y
horizontal and vertical components of Darcian velocity
~v = (v
x
; v
y
) Darcian velocity vector
v = v
x
i v
y
complex velocity
v
n
, v
s
normal and tangential (to the interface) velocity components
w = / + i w complex potential
x, y Cartesian physical coordinates
z = x + i y = qexp[ix] complex physical coordinate
ap, bp, cp, dp angles of the tilted interfaces
/ velocity potential
w stream function
f = n + i g auxiliary complex variable
Dike a soil-lled embankment constructed in Holland for preventing inundations
Piping soil erosion due to seepage, which begins from the exit point (tailwater slope) by dislodging, mobilising and
ushing away soil particles (starting from the nest fraction) and propagating upgradient to the upper pool
of the dam
Seepage heaving upward motion of a large soil volume resulting from a vertical component of the hydraulic gradient
Suffosion seepage caused migration of ne particles into the void spaces between larger particles
1
The photos attached illustrate the collapsed section of a small dam in Wadi Bani Kharus, Oman. The dam was swept away by the Gonu ood.
A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301 1287
and Materials/Corps of Engineers protocols (e.g. [8]) still leads to a sharp transition of conductivity from k
1,2
to one of the
lter, k
f
, across the seepage path. Numerous concerns were raised that the mentioned uniformity coefcients and their mod-
ications for adjacent layers, as purely textural indicators, do not address any seepage eld characteristics, e.g. hydraulic gra-
dients, which can vary both in magnitude and direction along and across the interfaces between porous media of different
composition. Moreover, the seepage eld can vary with time owing to a sudden exposure of the structure to extreme ood/
tide reservoir head H
1
(e.g., the Dutch St. Elizabeth-1421 and North Sea-1953 oods, American Katrina-2005, or Omani
Gonu-2007), gradual clogging of the lter or other seepage-induced re-texturalisation incidents/processes, which may befall
the dam. The analysis of failure occasions may require answering the question: was failure caused by heaving of a large vol-
ume of soil on the downslope of the earthwork (poor initial design of the whole structure due to the lack of a drain) or by a
focused piping due to clogging of an initially effective lter (poor design of the drain or its improper maintenance)? This calls
for a detailed analysis of the seepage eld in the whole soil volume of the dam (global scale) and close to the interfaces (local
scale).
Zonation of the dam body as in Fig. 1 reduces seepage and makes the slope of the tailwater reservoir less susceptible to
sliding, but creates sharp changes of the hydraulic gradient along internal interfaces between dam zones (e.g. CD in Fig. 1)
with potential piping (suffosion, see [9] for the nomenclature and recent review), which should be taken into account in
Oman and other Gulf countries, where most dam shells are made of poorly compacted, cohesionless materials with almost
no natural clay content.
In the design of the dam, even solely seepage-related criteria are often poorly compatible. Normally the upstream head in
Fig. 1 is xed and if the core is absolutely impervious (Q = 0) the downstream shell is in the safest seepage-wise conditions. A
saturated zone, however, builds up to BA and, consequently, increases the structure-enfeebling pore pressure in upstream
shoulder (prone to, for example, liquefaction of the dam part adjacent to the upper pool). Similarly, drains/diaphragms in
downstream shell lower the phreatic surface as compared with a homogeneous dam body (stability-wise, an unsaturated
soil in the dam body is preferable, i.e. the lower is the free surface the better). Additionally, draindiaphragm intercept
the ne particles travelling with the seepage ow (ideally, no seepage erosion should occur). On another hand, from com-
parison theorems (see [10]) the draindiaphragm always increase the Darcian velocities and Q, compared to no-drain/di-
phragm regimes and this jeopardizes the dam stability. On one hand, the core clay in Fig. 1 should have as low k
2
as
possible in order to impede the saturated seepage across it. On another hand, ne dispersing clays are erosion-prone and
seepage can dislodge too ne clay particles from the core, entrain them into downstream shell and toe drain with a gradual
clogging of the latter. Fine particles mobilized (bad phenomenon) by seepage from an upstream zone of the dam (core of
upstream shoulder) can be transported by ow to a downstream porous zone where they can self-heal piping channels
(good phenomenon). Seepage in heterogenous dam bodies/foundations and associated mechanical phenomena are so com-
plicated that post-failure litigations involve numerical and analytical mathematical models with often conicting outcomes
(see, e.g., [11]).
In the old models of Pavlovsky [12], Dachler [13], Davison and Rosenhead [14], Casagrande [15], Nelson-Skornyakov
[16], Aravin and Numerov [17,18]
2
seepage through embankments was typically analytically studied by assuming the whole
cross-section in Fig. 1 to be a homogeneous entity. Heterogeneities were taken into account in a simplistic way, by assuming the
boundaries of zones of high and low conductivity to be equipotentials, streamlines, or isobars that is mathematically sound if
the conductivity ratio of two adjacent zones is very high or very small. In this way, Polubarinova-Kochina [19], (further abbre-
viated as PK77) solved the dam problem by the theory of linear differential equations, paving a road to numerous applications
in other branches of applied mathematics [20,21]. The ow problem in the coreshells of Fig. 1 is more complicated than in a
homogeneous dam due to the presence of refraction boundaries (AB and CD). Moreover, even if seepage in the downstream shell
zone is considered without a full refraction conjugation with ow in the core (we shall do it below), then unlike PK77 neither
point D nor C
1
of the phreatic surface are known, i.e. we have a hanging free boundary.
2
See the scanned images of Russian sources as supplementary materials.
Filter and chimney
drain,k
f
Downstream
coarse shell, k
1
Fine
core,k
2
Upper pool
B C
H
2
H
1
Tailwater
Impermeable base
A
H
M
y
x

Toe
drain, k
f
C
1
D
L b
D
1
G
z
Upstream
coarse
shell, k
1
g
Fig. 1. Vertical cross-section of a physical plane of a zoned dam with a triangular core.
1288 A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301
In practical engineering, notwithstanding the availability of analytical solutions in homogeneous ow domains (e.g. trap-
ezoidal dam cross-sections with sloping tailwater and upper pool boundaries, PK77), most design manuals (e.g., [18,4]) rely
on simplied and even simplistic formulae, which treat seepage as a 1-D DupuitForchheimer process (e.g., the Casagrande
adjusted parabolas). This ignorance of more adequate 2-D models stems from the mathematical complexity of the PK77
nal solutions. Clearly, resorting to 1-D approximations completely circumvents the 2- (or even 3-) D problem of internal
erosion on zonal interfaces, already obviated by the PK77-type homogenisation of the dam body in 2-D analytical models.
The proliferation of numerical techniques in the 1960th (e.g., [22]), which mince any zoned seepage domain into
standard grids, helped, on one hand, in modeling heterogeneous dam sections, taking into consideration 2- and 3-D satu-
ratedunsaturated transient ow conditions in porous media undergoing intricate compaction, ageing, swelling, etc., but,
on another hand, had an insidious effect on the analytical approach, which was denigrated for its lack of versatility and abil-
ity to deliver real seepage patterns. The analytical techniques indeed stumbled on the necessity to tackle the realistic
features of seepage/medium, in particular, the refraction boundary conditions on the interfaces between different zones
in Fig. 1. Gradually, it became clear that the FDM-FEM codes despite their proclaimed superpower are caliginous in
describing the ne features of the ow eld just in the vicinity of interfaces, where the jumps in Darcian velocity are of par-
amount importance for, say, the analysis of drain clogging and its long-term ability to check the phreatic surface and pore
pressure. So, the numerical solutions of the dam problem still recur to analytical classics (see, e.g., [23]). We recall that in
heterogeneous aquifers the same Darcian seepage does not normally effectuate any structural collapses and, hence, FDE-FEM
meshing although blurring the ne scale refraction gives an allegedly acceptable large-scale picture of groundwater
motion. If contaminant transport rather than bulk ow is of concern, then, as Bear [24] emphasized, homogenisation is
not suitable because local heterogeneities cause plume ngering similar to concentrated erosion patterns in dikes.
The recent advances in the theory of boundary value problems of R-linear conjugation [2528] made possible solving a
broad spectrum of seepage refraction problems (e.g. [2932]) for harmonic elds, scabrous for analytical treatment on the
days of the evoked dam classics. In this paper, we revisit the old schemes studied by Davison, Kalinin and Mikhajlov and
tackle those elements/peculiarities of ow in Fig. 1, which were overlooked/mathematically untractable on the days of dam
giants.
2. Seepage through arbitrary triangular core: hodograph and conformal mappings
We assume steady-state Darcian seepage in a homogeneous isotropic triangle ABC (G
z
) in Fig. 1 (to be shown in more de-
tails as Fig. 2a). The unsaturated moisture movement in the shells outside ABC is neglected. Although H
1
in Fig. 1 does not
always rise to point A, the most common approach in analytical studies of seepage is to consider this extreme (most dan-
gerous) regime, which Nelson-Skornyakov [16] called the maximum one. If H
1
is less than what is shown in Fig. 1, a phre-
atic surface in the core appears, seepage gradients, pore pressure and Q drop (as compared with the maximum case, see
[10] for rigorous statements of the corresponding variational theorems), but mathematically the problem becomes prohib-
itively complicated (see PK77, pp. 5863). For the maximum regime with no free surface, classical books give a compen-
dium of analytical solutions: Nelson-Skornyakov [16] studied the case of H
2
= 0, Davison [33, pp.261266], investigated a
special case of b = c = 1/4 and H
2
0, and Mikhajlov (see PK77, pp. 288289) presented an approximate solution for
H
2
0 and arbitrary core slopes. PK77 mentioned that the general case in Fig. 1 can be obtained by conformal mappings
but reported no results.
We ll in the lacunae in the Davison and Mikhajlov results and present a full solution for the case of arbitrary H
2
0 and
arbitrary acute angles pb and pc,b + c > 1/2 (the latter inequality guarantees that a phreatic surface is not formed near point
A in Fig. 1). Most engineered cores are tall and narrow enough such that this inequality is satised. If the core apex A is above
H
1
in Fig. 1, then a phreatic surface appears in the core and the problem becomes exceedingly complex. The focus of our
study is ow near points B and C in Fig. 1 and we surmise that the phreatic surface near the tip A can be neglected.
C
u
v
G
V (1/2)
V=u+i v
-ik
2
-ik
2
/2
B M
A
D
(1/2)
A B M C
D D

=+ i
-1 0 m c
A
B C
D
iQ
H

w=+i
D
G
w
G

A
B C
M
k
2
H
D

Core of conductivity k
2
G
z
(a) (b)
(c) (d)
Fig. 2. Flow through a triangular core. Physical domain (a), velocity hodograph pentagon (b), auxiliary half-lane (c), and complex potential domain (d).
A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301 1289
The objective of this section is to determine the seepage eld as w - z in a triangular core. For this purpose we use the
hodograph method. We introduce the complex physical plane z = x + i y with the origin at point B and show the core in Fig. 2a
which represents the corresponding dam element in Fig. 1 but without any lter/chimney or toe drain. The hydraulic head
h(x, y) obeys the Laplace equation in ABC. The head drop from AB to CD is H = H
1
H
2
, i.e. the reference level is H
1
(tailwater).
In this section CD is a constant head boundary. The segment AD of the core is a seepage face along which the pressure head
(dened by p = h y) is zero, i.e. h = y. The foundation BMC is horizontal and impermeable. If the water level in the upper
pool (left of the core in Fig. 2a) drops, then a phreatic surface appears and then one will be on the safe side. Indeed, from
the comparison theorems (see, e.g. [47]) follows that the head drop along AB results in redcution of the size of the seepage
face, smaller pore pressure and Darcian velocities inside G
z
in Fig. 2a.
The Darcian velocity vector ~v = k
2
\h and complex potential w(z) = /(x, y) + i w(x, y), / = k
2
h are introduced as in PK77.
The hodograph domain G
V
is a pentagon shown in Fig. 2b where the cut tip, point M, corresponds to the maximum of Darcian
velocity along BC. In G
V
AB is normal to the boundary AB in the physical plane, and both CD and AD are normal to AC in G
z
(see
[33], and PK77). The complex potential domain G
w
is shown in Fig. 2c where the thin line AD is a curve, whose shape is not
known in advance.
Solution is obtained by conformal mapping of G
z
and G
V
onto an auxiliary half-plane g > 0 (G
f
) of a variable f = n + i g
(Fig. 2d), where m and c are the afxes found from solving a set of nonlinear equations (the details are given in Appendix A).
Example 1. For b = 0.3, c = 0.4, L = 10 m, H
2
= 2 m and H = .75 m the parameters c = 7.76, m = 0.317, Q/(k
2
L) = 0.462.
Our solution is expressed as z(f) and dw/dz(f) via an auxiliary variable f involved in the hypergeometric function
2
F
1
. On
the days of Davison and PolubarinvoaKochina tackling these functions, including their integration, was technically cumber-
some. Now days, computer algebra packages have special functions as built-in routines and analytical solutions, based on the
theory of holomorphic functions (e.g. mapping of a z-triangle onto a w-pentagon in Fig. 2), are invigorated.
2.1. Primitive conjugation of seepage in core and downstream shell
In the above presented solution H
2
was specied by the level in the tailwater and the only seepage-resisting element of
the dam was its core. In this subsection we consider the core conjugated with a downstream shell as shown in Fig. 3 and H
2
obtained fromsolution. We consider a toe drain, which forestalls exltration through the right slope of the damand no chim-
ney drain. The core in Fig. 3 is relatively ne in texture and rectangular in shape. The distance between the right face of the
core and the drain is b.
The branch AD of a phreatic surface in the core is disconnected from the branch D
1
C
1
in the downstreamshell with a seep-
age face emerging on the dowstream side of the core. We assume that all water, which passed through the core and entered
downstream shell (both through the seepage face and through the constant head segment D
1
C in Fig. 3), is collected in a sat-
urated zone below D
1
C
1
in Fig. 3. All this water is intercepted by the drain, albeit moisture exuded from the core through DD
1
can temporarily loop through the unsaturated zone of downstream shell prior to joining the saturated zone through D
1
C
1
.
Head loss in the core and downstream coarse shell are H and H
2
, respectively.
According to the Dupuit formula the ow rate through a rectangular dam with water levels H
1
and H
2
is
Q = k
2
H
2
1
H
2
2
2L
; (1)
that is an exact value in both the DupuitForchheimer approximation and full 2-D theory, as Charny showed (see PK77).
The downstream shoulder in Fig. 3 is usually relatively broad (b H
2
). Consequently, we can use the DupuitForchhei-
mer approximation for ow in this coarse zone, in congruity with which
Seepage in the fine
rectangular core
Upper pool
B C
H
2
H
1
Impermeable base
A
H
x
Toe drain
C
1
D
L b
D
1
Seepage in the
coarse downstream
shell
G
z
G
d
b
d
g
k
2 k
1
Phreatic surface
in the core
Phreatic surface
in the downstream
shell
Seepage face of
the core
Fig. 3. Physical plane of a dam with a rectangular core and diaphragm.
1290 A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301
Q = k
1
H
2
2
2b
; (2)
where we assumed that the water level in the toe drain is zero. If the upstream face of the toe drain is vertical then (2) is
exact.
We note that an exact but more cumbersome (i.e. not relying on the DupuitForchheimer theory) solution for ow in the
downstream shell is given by PK77 for arbitrary slopes of CD and the drain trench.
Combination of (1) and (2) yields
H
2
H
1
=

k
r
k
r
L
r

; k
r
=
k
2
k
1
; L
r
=
L
b
: (3)
Example 2. For H
1
= 7 m, k
1
= 10 m/day, k
2
= 0.1 m/day, L = 3 m, b = 10 m from (3) H
2
= 1.26 m.
A similar conjugation of the core and shell ows can be carried out for the triangular core of the previous section but the
result is not as simple and explicit as in (3) for a rectangle.
We note that despite its simplicity the conjugation of this kind has not been carried out in the standard manuals (e.g.
[18]) where the ignorance of the loss of head in downstream shell for the scheme in Fig. 3 would imply H
2
= 0, which is, obvi-
ously, not acceptable if seepage-induced sliding in downstream shoulder is of concern.
3. Kalinins wedge as a generic refraction element
In this section we scrutinize the Kalinin [34] solution to the problem of seepage in a two-layered wedge (Fig. 4a).
Although Fig. 4a shows the upstream shoulder on the left and core on the right of an interface OB, there is no limitation
on k
1
, k
2
i.e. Fig. 4a can represent any coreshelllterliner-facing conjugation. In this section we adopted the Kalinin nota-
tions but permuted his notations as k
1
?k
2
.
The ray AO in Fig. 4 is at a constant head, which here we assume to be zero. The face ONC of the core is at atmospheric
pressure i.e. h = y along this ray. Geotechnically, this boundary condition is ensured by a chimney drain/lter, which inter-
cepts water and clay particles [5]. Along OB we have a refraction of the streamlines. One of which, EMN, is shown swerving at
point M (all other streamlines are homothetical to EMN). We assume that both wedges, AOB (S
1
) and BOC (S
2
) are completely
saturated (i.e. no phreatic line appears there). The effect of the dam base BC (Fig. 1) is ignored i.e. points A, B, and C in Fig. 4a
are at innity. The angles ap, bp and cp in Fig. 4a represent the wedge sizes and orientation with respect to the gravitational
eld. The Darcian velocities in the core, ~v
1
, and in the upstream shell, ~v
2
, are constant.
We re-derived Kalinins [34] results for ow in Fig. 4a (see Appendix B) as
V
2
=
[
~
v
2
[
k
2
=
K
r
cos pc cos pb
1
K
r
sinpa cos pb
1
tanpb cos p(a b
1
)
;
V
1
=
[
~
v
1
[
k
1
=
V
2
cos pb
1
=(K
r
cos pb)
; K
r
= k
1
=k
2
;
(4)
yy
O
Interface y
x
O y
Equipotential
E i t ti l li
V
1 x
O
q p
lines in
Equipotential line
g
1 O lines in
medium 1

g
k
medium 1

k
2

Chimney
k
1

Chimney
drain/filter
1
drain/filter
k
V
2
k
1
E
2
E i i l
Seepage
E Equipotential
k
2
p g
face
/2
lines in
2 face
V

/2
medium 2
Isobars
V
1

p=const
M
/2
p
in medium 2 Seepage
/2
Seepage
face p=0
A V
2
face p=0
in medium 2
S
1
A V
2

1
in medium 2
S
1

1
Refracted
streamline
S B
streamline
S
2
B
N
CC
(a) (b)
Fig. 4. Vertical cross-section of the Kalinin two-layered wedge (a) and the corresponding full-plane refraction pattern with a straight interface (b).
A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301 1291
~v
1
is perpendicular to the equipotential AO and the orientation of ~v
2
is determined by the refraction condition along BO:
K
r
= tanpb= tanpb
1
; (5)
where b
1
is the angle that ~v
2
makes with the normal to BMO. Kalinin imposed the following no-phreatic-surface condition
b
1
6 1=2 a: (6)
An additional inequality:
V
1
< cos p(a c)= sinpb; (7)
should be satised. Kalinin missed (7), which stipulates that ow on the lee-side of a low-permeable barrier should remain
saturated i.e. pore pressure to remain positive (see Appendix B). Physically, in a relatively thick upstream shoulder or chan-
nel siltclay liner [35] the pressure head of ow passed through a low-permeable zone of sufcient thickness can become
negative. Then what is shown as refraction of EM in Fig. 4a is not true. Consequently, the unsaturated ow in k
2
zone of
Fig. 4a becomes an inltration shower (term used in PK77). In this case (possible for k
2
> k
1
only, see [36] for more details)
(7) does not hold. It is noteworthy that a cascade of intermittent saturatedunsaturated zones can emerge in intermittently
heterogeneous porous media [37].
Fig. 5a depicts V
1
and V
2
as functions of b for K
r
= 0.3, a = 0.3, c = 0.1. This case corresponds to core- downstream shell
conjugation (K
r
< 1). The upper and lower curves in Fig. 5a correspond to the value of hydraulic gradient in the core (left
wedge) and downstream shell (right wedge) in Fig. 4, correspondingly. The vertical line cuts the solution (4) according to
(7), without which Kalinins [34] solution would extend to b ~ 0.065, if limited solely by (6). In Fig. 5b the gradients are plot-
ted as functions of a for K
r
= 3, b = 0.2, c = 0.1. Recurring to Fig. 4 this case corresponds to the conjugation between upstream
shoulder and core. The upper and lower curves in Fig. 5b show V
1
and V
2
in upstream shell (now the right wedge in Fig. 4)
and the core (the left wedge), correspondingly. For K
r
< 1 the most dangerous (suffosion-wise) segment is the ray OB, i.e. V
1
in
Fig. 5a is critical. For K
r
> 1 (upstream shoulder-core/chimney drain conjugation), the segment OC in Fig. 4, (i.e. V
2
from
Fig. 5b) is critical for assessments of piping.
Similarly to the Davison seepage scheme and in the sense of variational theorems discussed in the Introduction, the full-
saturation ow shown in Fig. 4 is most dangerous, i.e. in case of an emerging unsaturated zone on the right of OB the pore
pressure and ow rates will drop and structural stability of the composite wedge of Fig. 4 improves.
4. General solution for refraction on a wedge
In this section we consider those elements of the dam in Fig. 1, which are characterized by so-called triple points where
two adjacent porous zones are underlaid by an impermeable base. Near these points (B, C and C
1
in Fig. 1) the Darcian veloc-
ity vector has singularities if the angles b and c are unfavourable (we will specify later the deleterious slopes). Then
Fig. 5. Hydraulic gradients in the two components of the Kalinin wedge as functions of b for K
r
= 0.3, a = 0.3, c = 0.1 (a) and as functions of a for K
r
= 3,
b = 0.2, c = 0.1 (b).
1292 A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301
suffosion takes place in the zones where the hydraulic gradient spikes above a certain limit. PK, 1977 recommends 1 as this
critical gradient, although other (higher and lower) empiric limits are used in geotechnical engineering.
This local increase of the gradient magnitude is caused by the impermeability or low permeability of the foundation (we
recall that below the line BCC
1
in Fig. 1 the soil is impervious) as has been recently described by Fox and Wilson [38]. Indeed,
an impermeable base prevents heaving but facilitates localization of high-gradient outseeps. For the sake of deniteness, we
consider the conjugation of upstream shoulder and core shown in Fig. 6 corresponding to b > 1/2 in Fig. 1 (if b < 1/2 or equiv-
alently d > 1/2 in Fig. 6 then the gradient at point B is zero and the vicinity of this point is erosion-wise safe).
Here we study the vicinity of point B (Fig. 6), which represents a zoomed area of the corresponding point in Fig. 1. Math-
ematically identical seepage refraction arises at triple points like B in Figs. 1 and 6 of conjugated coredownstream shell,
diaphragmdownstream shell, ltercore, lterdownstream shell, and downstream shellriprap (see, for example, points
C and C
1
in Figs. 1 and 3). Our ow domain in Fig. 6 is bounded by curved lines MN and NC. It would be better to consider
two adjacent triangles: MNB and BNC. This geometry is, however, mathematically intractable and reconstruction of MN and
NC as parts of solution is and inverse trick (see [16]).
In the two previous sections we conjugated seepage in the core and shell through either a simple mass-balance condition
(outow from one homogeneous element of the ow domain equals inow into an adjacent domain of contrasting conduc-
tivity) or by matching two 1-D ows in two wedges (Kalinins scheme with a rigorous conjugation of velocities and heads
along a ray). Now we shall study a fully-coupled, 2-D Darcian ows in the angles ABA
u
and EBE
u
separated by an interface
A
u
NBNE
u
.
We assume that the seepage domain in Fig. 6 is bounded from below by an impermeable base, the horizon AMBCE. We
emphasize that if the twosupplementaryangles ABA
u
(domainS
1
) andEBE
u
(domainS
2
) were made of the same porous medium
and are considered separately, then their innities would be mapped on one point for each Riemann sphere (PK77). This
means that in a homogeneous medium points A, A
u
, E
u
and E coincide. In Fig. 6 the innities of the two angles are different.
The interface BNA
u
(E
u
) is a ray making an angle dp with BA of the dam base. The origin of the (xy) Cartesian coordinates is
now at point B. In S
2
we have a fully saturated Darcian ow with the complex potentials w
2
and velocity ~v
2
= (v
2x
; v
2y
) and in
S
1
the corresponding functions are w
1
and ~v
1
. w
1
and w
2
are holomorphic in S
1
and S
2
. Along the interface we have a standard
refraction conditions (PK77):
v
1n
= v
2n
; v
1s
=k
1
= v
2s
=k
2
; (8)
where v
1n
, v
2n
are the components of
~
v normal to A(E)B and v
1s
, v
2s
are tangential components. Generalizations of (8) to the
case of anisotropic media and two-phase ows are given by Mualem [39], Raats [40].
The objective of this section is to nd w
1
and w
2
as explicit functions of z for an arbitrary slope d of a straight interface in
Fig. 6 and to interpret these two conjugated holomorphic functions in a geotechnically meaningful manners. For these pur-
pose we select the classes of w
1
and w
2
that are either nite or innite at point B and at innity. We note that these re-
mote points A in S
1
and E in S
2
are different in the two zones (by contrast, in a homogeneous soil they would be imaged by
the same point on the Riemann sphere).
The character of singularity at point B can be mathematically different. Grinberg [41] solved a mathematically equivalent
problem in electrostatics and in the class of two holomorphic functions with an essential (non-integrable) singularity at
point B (see his Chapter XIV, Section 38). We restrict ourselves by a narrower class of integrable singularities at B. For exam-
ple, a logarithmic singularity for w
1
,w
2
at B, is usually interpreted as a drain (well) according to the well-known (PK77) rep-
resentation of a line 2-D sink or source. Clearly, for geotechnical applications we cut the corresponding vicinities of B.
Obnosov [42] found two holomorphic functions (complexied velocities) v
j
(z) = v
jx
i v
jy
= dw
j
/dz, j = 1, 2 in the two do-
mains S
1
and S
2
of Fig. 6 as described below. At d (0, 1/2) the general solution is:
v
1
(z) =

N
j=1
c
j
K(h
j
)(z)
h
j
1
; z S
1
;
v
2
(z) =

N
j=1
c
j
z
h
j
1
; z S
2
;
(9)
k
2
k
1
A B
Impermeable base
A
u
y
x
E

N
M C
Upper
pool
H
1
Seepage face
Constant-head line
Interface
S
1
S
2
N
1
z=x+iy
E
u
g
L
L
1
M
1
M
2
Fig. 6. Vertical cross-section of an upstream shell conjugated with core through a tilted interface.
A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301 1293
where c
j
, j = 1, 2. . . , n, are arbitrary real parameters, and h
j
, j = 0, 1, 2, . . . , N, are any N roots of the equation
sin(ph) Dsin[p(1 2d)h[ = 0; (10)
where
D = (k
1
k
2
)=(k
1
k
2
):
In (9) the single-valued branches of the functions z
h
j
and (z)
h
j
are xed in the domains S
2
and S
1
by the condition that these
functions are real at z = x > 0 and z = x < 0, correspondingly. The function K(h) is dened as:
K(h) =
sin[p(1 d)h[
sin(pdh)
=
k
1
cos[p(1 d)h[
k
2
cos(pdh)
; (11)
Elementary inspection shows that Eq. (10) has an innite number of roots (both positive and negative) which condense at
innity. Each root corresponds to one term in (9). In (9) we retained non-negative roots only.
In what follows we consider two special cases of truncation of (9), viz. the so-called one-term solutions. Without any loss
of generality we assume k
1
> k
2
and 0 < d < 1/2. Obviously, for d > 1/2 in the above-written solution one has to permute
v
1
?v
2
.
5. Drain/lter centered at the triple point
As we discussed above, near points B, C and C
1
the porous continua of three different conductivities (k
1
, k
2
and 0) are most
susceptible to suffosion. Mitigation of internal erosion (translocation of soil particles from one zone to another and an ensu-
ing decrease of structural stability) is achieved by constructing a drain/lter diagrammed in Fig. 7. Here we consider a semi-
circular lter of a radius R centered at point B. Water seeps into the lter from both wedges (S
1
and S
2
) such that the lter
contour MNC (an interface between the gravel pack inside and the interior wedge media) is a constant-head line. Solution to
this problem follows immediately from (9) if we retain a single term corresponding to the root h
0
= 0, K= k
1
/k
2
. Then
v
1
(z) = c
0
k
1
=(k
2
z); z S
1
; v
2
(z) = c
0
=z; z S
2
: (12)
It is well-known (PK77) that a point sink in 2-D ow is determined by v = c
0
/z where c
0
is an arbitrary real constant and v(z) is
a Darcian velocity eld in a homogeneous porous plane. Our solution (12) is of the same sink type but for two adjacent
wedges. Indeed, we select w
2
= 0 along BCE and /
2
= 0 along the circle NC. Then the two complex potentials in (12) corre-
spond to a drain placed at the same point B such that the head is counted from the drain contour (NC). Along this semicircle,
we have z = Rexp[ix], 0 < x< p(1 c), where x is the angular coordinate. We integrate the second equation in (12) that
with the selected references for the potential and stream function gives in polar coordinates:
w
2
(z) = /
2
iw
2
= c
0
log
q
R
ic
0
x; z S
2
; (13)
where q is the radial coordinate in the complex plane.
Integration of the rst equation in (12) yields:
w
1
(z) = /
1
iw
1
= c
0
k
1
=k
2
log
q
R
U
0
iW
0
; z S
1
: (14)
The real part, U
0
, of the constant of integration is determined from the condition that /
1
= 0 along MN (the segment of our
semicircle in the left wedge S
1
) because inside the gravel pack water is in hydraulically identical conditions, independent of
whether it touches the rst or second medium in the exterior. The imaginary part, W
0
, follows from the condition w
1
= w
2
at
point N. Actually w
1
and w
2
coincide along the whole interface ray NA (NE). Consequently,
w
1
(z) = c
0
k
1
k
2
ln
q
R
i c
0
p
k
1
k
2
(1 x) (1 d)
k
2
k
1
k
2
_ _
; z S
1
: (15)
A Interface (stream line) E
Constant
y
Constant-head
semi y semi-circle
SS
2
k
2 z=x+iy 2
N
S
z=x+iy
R
k
1
S
1
1

x
A B
x
E M C A B
Impermeable base
E M C
Filter filling with k
f
>>k
1
, k
2 Line sink f 1
,
2
R

Fig. 7. Vertical cross-section of a circular drain centered at a triple point.
1294 A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301
From (15) at point M we have
w
1M
= Q = c
0
p(1 d)
k
1
k
2
k
2
: (16)
Here Q is the total seepage inow into our semicircular lter. From (16) we express the constant c
0
through Q. Then the in-
ow through CN is
w
2N
=
Q
p
k
2
k
1
k
2
:
(we recall that in (16) we assumed k
1
> k
2
and 0 < d < 1/2).
This almost trivial solution, characterized by a purely radial inow into the drain with no uid motion across the interface
between S
1
and S
2
, will be used in the next section as a mode of inverse shaping of the domain boundaries. It is noteworthy
that the Kalinin [34] scheme in Fig. 4a can be obtained (by a similar inverse procedure) from Obnosov [42] who studied an
arbitrary orientation of a two-layered wedge with respect to an arbitrary ow. Indeed, in Fig. 4b we are showing a standard
(e.g. [43]) problemof refraction on a straight interface. The equipotential lines (dashed lines in Fig. 4b) on the left and right of
the interface are straight. Any of these lines in the rst medium can be selected as AO in Fig. 4a. The isobars (solid lines in
Fig. 4b) are also straight. Therefore, by selecting one passing through point O in Fig. 4b and setting pressure to be zero on this
ray we obtain the physical interpretation of the coreshell conjugation in the Kalinin scheme of Fig. 4a.
6. Shellcore conjugation
Now we come back to Fig. 6 and consider another special case of (9). We formally retain in the series one term, which
corresponds to the root of Eq. (10) on the interval 0 < h < 2, and will see what is the physical scheme of seepage correspond-
ing to this particular case. For positive D Eq. (10) has one and only one root in the selected interval of h. This root satises the
inequality 0 < h
1
< 1. Obviously, the value of the root depends on d and the ratio k
1
/k
2
. The corresponding one-term trunca-
tion of (9) is:
v
1
(z) = K(h
1
)c
1
(z)
h
1
1
; z S
1
;
v
2
(z) = c
1
z
h
1
1
; z S
2
;
(17)
where K(h) is dened in (11) and c
1
is a real parameter to be determined later.
In particular for d = 1/4 from (9)(11) we get
v
1
(z) = (1 D)c
1
(z)
h
1
1
; z S
1
;
v
2
(z) = c
1
z
h
1
1
; z S
2
;
(18)
where h
1
= (2/p) arccos(D/2), 0 < h
1
< 1.
Obviously, at d = 1/2 the wedges degenerate into two adjacent half-planes and the corresponding trivial solution follows
from (9) as:
v
1
(z) = v
2
(z) = c
1
;
i.e. we obtain the known [43] unidirectional ow perpendicular to an interface.
Now we utilize (17) for reconstructing a nite-size domain MNCBM consisting of a ne dam core and coarse upstream
shell. We implement an inverse approach of Nelson-Skornyakov [16] resuscitated by Obnosov [44]. The idea of this ap-
proach is to conjugate in a rigorous manner two complex potentials (or complex Darcian velocities) along a given curve
(an interface between two media of contrasting conductivity, e.g. parabola as in Kacimov and Obnosov [44]). The two ana-
lytic functions will have certain singularities at given isolated points. Then certain streamlines, equipotentials or isobars of
these functions are interpreted as boundaries (stream tubes) of seepage domains. Here we reconstruct a constant-head
boundary MN and an isobar (seepage face) NC (Fig. 6) (again, (17) is exactly satised!). We notice that Nelson-Skornyakov
[16] himself used this inverse procedure for ows in homogeneous media only.
Pressure head in the two domains is (see PK77)
p
1
=
/
1
k
1
y c
p
_ _
; p
2
=
/
2
k
2
y c
p
_ _
; (19)
where the real constant c
p
will be determined later.
Integration of (17) results in two complex potentials
w
1
(z) = /
1
i w
1
= c
1
h
1
1
K(h
1
)(z)
h
1
; z S
1
;
w
2
(z) = /
2
i w
2
= c
1
h
1
1
z
h
1
; z S
2
:
(20)
The constants of integration are zero in (20). This follows from adopting two reference values. First, at point B we set
w
1
= w
2
= 0 because both AMB and BCE are streamlines and unlike the previous section there is no mass loss (generation)
A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301 1295
at point B. Second, at this point we set /
1
= /
2
= 0 because the velocity potentials in both media are determined up to a real
constant. We emphasize the difference between (18), (20) and the solution from the previous section where the drain solu-
tion at z = 0 has a logarithmic singularity for both w
1
and w
2
. We isolated this singularity by excluding a semi-circular vicin-
ity of point B (Fig. 7), interpreted as the contour of lter lling. The complex potentials in (20) at z = 0 are nite (zeros),
although the velocity is not. The singularity of velocity in (18) is said to be integrable (PK77). Unfortunately, unlike the pre-
vious section, Eq. (20) do not give simple known curves as equipotentials in S
1
and S
2
. Let us nd out what are these curves.
If we use polar coordinates with the origin at B in Fig. 6, then z = qexp[i xp] where xp is the angle counted from Bx. In S
2
and S
1
we have 0 < x< 1 d and 1 d < x< 1, correspondingly. From (20) the potentials are:
/
1
= c
1
K(h
1
)h
1
1
q
h
1
cos[ph
1
(x1)[; z S
1
;
/
2
= c
1
q
h
1
h
1
1
cos(ph
1
x); z S
2
(21)
and the stream functions are
w
1
= c
1
K(h
1
)h
1
1
q
h
1
sin[ph
1
(x1)[; z S
1
;
w
2
= c
1
h
1
1
q
h
1
sin(ph
1
x); z S
2
:
(22)
The general solution is dened in an innite domain S
1
S
2
, i.e. in the whole half-plane. As usually [45], piezometric data
should be incorporated for specication of the incident seepage ow. Here we do it by specifying the head value at point
M, i.e. /
M
= /
0
, (/
0
> 0). This point is located distance L from B (x
M
= q
M
= L). Then from the rst Eq. (21) we obtain
c
1
= /
0
h
1
L
h
1
=K(h
1
):
Consequently, (21) and (22) become:
/
1
= /
0
(q=L)
h
1
cos[ph
1
(x1)[; q > 0; 1 d < x < 1;
/
2
= /
0
=K(h
1
)(q=L)
h
1
cos ph
1
x; q > 0; 0 < x < 1 d
(23)
and
w
1
= /
0
(q=L)
h
1
sin[ph
1
(x1)[; q > 0; 1 d < x < 1;
w
2
= /
0
=K(h
1
)(q=L)
h
1
sinph
1
x; q > 0; 0 < x < 1 d:
(24)
Now we reconstruct an equipotential / = /
0
, which corresponds to NM in Fig. 6. This line is the inlet of our ow
domain and represents the interface between the reservoir (open water body) and a porous upstream shell. H
1
is the reser-
voir water level above the impermeable bed. Therefore, at point M the pressure head is H
1
and from the rst Eq. (19) we
obtain
c
p
= H
1
/
0
=k
1
: (25)
The value of H
1
is found from the condition that at point N pressure is zero because at this point the free water in the res-
ervoir contacts both the upstream shell and atmosphere. Therefore, y = H
1
, q
N
= H
1
/sin[pd], / = /
0
. The straight interface BN
is a refraction line and hence k
2
/
1
= k
1
/
2
there (PK77). In particular, /
2N
= (k
2
/k
1
)/
0
at the point N. From the second Eq. (23)
and the relation (11) we get
/
2N
= /
0
(q
N
=L)
h
1
cos[ph
1
(1 d)[=K(h
1
) = (k
2
=k
1
)/
0
(q
N
=L)
h
1
cos[phd[:
From these two relations we nd
H
1
= L
sin[pd[
[cos(pdh
1
)[
1=h
1
; q
N
=
L
[cos(pdh
1
)[
1=h
1
: (26)
From Eqs. (25), (26) we express c
p
.
While in S
1
we have just reconstructed the reservoirupstream shell interface as an equipotential, in S
2
we will do the
same but for a seepage face NC. Why not a standard stream tube (two equipotentials and two streamlines)? The answer
is: we need a owdomain mimicking the damcore in Fig. 1 or Fig. 4 (and in the original Kalinin [34] owpattern in a wedge).
Through an isobar CN all water inltrated into the upstream shell and refracted on BN is discharged. The fate of water after
leaving CN is not important for us, this water can exude into downstream shell or chimney drain not shown in Fig. 6. From
the second Eq. (21)
/
2
=k
2
y H
1
/
0
=k
1
= 0: (27)
From (27) using the second Eqs. (23) and (26) we obtain the polar equation of NC:
/
0
K(h
1
)k
2
(q=L)
h
1
cos(pxh
1
) qsin(px) =
L sin(pd)
[cos(pdh
1
)[
1=h
1

/
0
k
1
: (28)
This equation was solved by the FindRoot routine of Mathematica.
1296 A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301
From the second Eqs. (24) and (26) follows
w
1N
= Q = /
0
tan(ph
1
d):
At last, from the second Eqs. (19) and (25) follows that /
2C
= k
2
(H
1
+ /
0
/k
1
) at point C (we recall that CN is a seepage face
with p
2
= 0). At the same point we have /
2C
= /
0
=K(h
1
)(q
C
=L)
h
1
due to (23). Wherefrom, using the relation (26), we get
q
C
= x
C
= L
1
= L
k
2
K(h
1
)
/
0
L sin[pd[
[cos(pdh
1
)[
1=h
1

/
0
k
1
_ _ _ _
1=h
1
:
Fig. 8 shows the physical domain in dimensionless coordinates (X,Y) = (x/L, y/L) for k
2
/k
1
= 0.2, d = 0.4 and /
0
/(k
1
L) = 1.
Lines 15 are the interface, MN,NN
1
, the streamline w
1
~ 2.0 coming to point N from the rst medium, and NC (correspond-
ingly). Curves 2, 4, and 5 extend to the adjacent medium (i.e. go beyond the intersection point N), where, of course, the
equipotentials, streamlines and isobars obey different equations. Fig. 9 shows the refraction of the streamlines
w
1,2
/(k
1
L) = 2.0, 1.5, 1.0, 0.5 (curves 14) on the interface. Fig. 10 shows the contours of isotachs V
2
= 4, 3.8, 3.6 (curves 1
3, correspondingly). These contours evince point-wise and global criteria of stability of soil slopes [46]. Fig. 11ac shows
H
1
/L, L
1
/(k
1
L) and Q/(k
1
L) as functions of k
r
= k
2
/k
1
for d = 0.45, 0.35, 0.25 (curves 13, correspondingly).
To summarize this most difcult section of the paper we can extend the Nelson-Skornyakov type reconstruction tech-
nique to other than (20) pairs of refraction-conjugated complex potentials. The selected solution (20) is in reality very sim-
ple: to generate the core and shell shapes we need actually one point M in Fig. 6 (and of course conductivities and the slope
of the interface). For more terms retained in (9) the reconstruction procedure may be more sophisticated.
- 2 - 1 0 1 2
0
0.5
1
1.5
2
2.5
3
3.5
1 2
3
4
5 X
Y
Fig. 8. Reconstructed shell-core for k
2
/k
1
= 0.2, d = 0.4 and /
0
/(k
1
L) = 1. Lines 15 are the interface, MN, NN
1
of Fig. 6, the streamline w
1
~ 2.0 coming to point
N from the rst medium, and NC.
- 2 - 1 0 1 2
0
0.5
1
1.5
2
2.5
3
3.5
1
2
3
4
X
Y
Fig. 9. Streamlines w
1,2
/(k
1
L) = 2.0, 1.5, 1.0, 0.5 (curves 14) for the dam in Fig. 8.
A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301 1297
What is the practical value of obtaining boundaries that are unlikely to be practically constructed of exactly the shape
shown in Figs. 10 and 11, or to appear like these curves in nature? This kind of solution can be used as upper and lower
-2 -1.5 -1 -0.5 0 0.5 1
0.5
1
1.5
2
2.5
3
3
1
2
seepage face
X
Y
Fig. 10. Isotachs V
2
= 4, 3.8, 3.6 (curves 13) for the dam in Fig. 8.
0.2 0.4 0.6 0.8 1
kr
1
2
3
4
5
6
7
H
1L
2
3
1
0.2 0.4 0.6 0.8 1
k
r
2
4
6
L
1
kL
2
3
1
0.2 0.4 0.6 0.8 1
k
r
1
2
3
4
5
6
7
Q kL
2
3
1
Fig. 11. H
1
/L (a), L
1
/(k
1
L) (b) and Q/(k
1
L) (c) as functions of k
r
= k
2
/k
1
for d = 0.45, 0.35, 0.25 (curves 13).
1298 A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301
bounds in the sense of variational theorems (see [10,47]). For example, let the inversely obtained boundary MN in Fig. 6 is
sandwiched between MN
1
and MN
2
as shown by dashed lines. Then these theorems stipulate that the ow characteristics
are also sandwiched. In particular, the total ow rate through the two-layered soil is the highest for the straight upper face
M
1
N and the lowest for M
2
N. We repeat that straight faces for both MN and NC would be most interesting to study from an
engineering viewpoint but mathematically this problem is nasty for an analytical treatment. What we reconstruct as curved
domain boundaries is a surrogate, useful in estimates.
The general solution (8) with only one term in the series retained gave two interesting ow schemes a sink acting in a
wedge-type heterogeneous half-plane and a constant head seepage face ow tube with an internal tilted interface. We sur-
mise that with more terms retained in (8) other interesting seepage schemes can emerge.
7. Conclusion
Mathematically, potential elds in heterogeneous media are tractable by analytical techniques if a piece-wise constant
approximation of conductivity is adopted this is often justiable in articially zoned dam bodies and graded lters. Then
at the interface between two porous zones two holomorphic functions must satisfy the conditions of a homogeneous general-
ized Riemann boundary-value problem (the problem of R-linear conjugation). Generally speaking, an arbitrary problem of R-
linear conjugation cannot be solved analytically in a closed form. Only few specic heterogeneities admit an explicit solution.
The general refraction problem(8) for hyperbolic, (wedge in Fig. 4 is a limit case of a hyperbola), elliptical and parabolic inter-
faces in the most general class of arbitrarily distributed singularities has been solved in [26,27]. These new solutions - often
seemingly exotic - enhance the arsenal of analytical modelers, especially those, who implement the Analytical Element Meth-
od of Strack [48], which requires (for bench-marking) simple closed-form solutions. For example, Anderson [49] utilized the
PK77 solution for a well with a formation damage in a very different hydrological problem of groundwater-surface water
interaction. Similarly, the analytical solution (9) in this paper exploited for civil engineering applications can be imple-
mented not only in subsurface mechanics of Darcian ows but also in electrostatics, steady heat conduction and diffusion
(e.g., [41]).
On the days when the theoretical foundation of dam classics was laid the predicted seepage eld could be monitored
through piezometer networks in the dam body, which give point values of the pressure head only. Modern technologies (see,
e.g., [50]) brought to practice novel tools, which can measure the magnitudes and even (to some degree) the directions of the
Darcian velocities that is promising for comparing analytical solutions written in terms of velocity.
The presented results serve the goal of catching breaches in numerico-hoaxes and acquiring a better insight of focused
seepage as a precursor of geotechnical breaches in hydraulic structures.
Acknowledgments
This work was supported by DOPSAR project Effects of siltation behind Al-Khod dam on the soil properties and recharge
efciency (Oman) and RFBR Grant No. 09-01-97008-r_povolghe_a (Russia). Helpful comments by two anonymous referees
are appreciated.
Appendix A
The rst conformal mapping of ABC (Fig. 1) onto G
f
= {f : Imf > 0} (Fig. 2b) is given by the SchwarzChristoffel formula
with the correspondence of points A ?1, B ?0, C ?c as
z(f) = a
1
_
f
0
(1 s)
bc
(c s)
c1
s
b1
ds; (29)
where the branch of the multivalued function (1 + f)
bc
(c f)
c1
f
b1
is xed in G
f
as positive at f = n, 0 < n < c.
The two mapping parameters in (29), a
1
and c (Fig. 2b) are found from the given sizes of the triangle G
z
in Fig. 1 and the
selected image of point D (D ? in the f plane) as following. At f = c we have z = L. Therefore, from (29)
a
1
=
L
I
1
; I
1
=
_
c
0
(1 s)
bc
(c s)
c1
s
b1
ds =
c
cb1
(c 1)
b
B(b; c); (30)
where B(x, y) is the Beta-function. Next, at f ? we have z
D
= L H
2
cotpc + i H
2
. Calculating the integral by the Integrate
routine of Wolframs [51] Mathematica we obtain
H
2
=L = I
2
sin(pc)=I
1
; I
2
=
_

c
(1 s)
bc
(s c)
c1
s
b1
ds =
B(1 b c; c)
c
1bc
(1 c)
b

F(1; 1 b; 2 b c; c=(1 c))
(1 c)(1 b c)
; (31)
where F =
2
F
1
is the hypergeometric function [52]. From the nonlinear Eq. (31) we remembering I
1
from (30) calculate c
using the routine FindRoot of Mathematica.
The function V = v
x
+ i v
y
is antiholomorphic and we get rst a domain symmetrical with G
V
(Fig. 2a) about the abscissa
axis. The conformal mapping of this mirrored domain (corresponding to the holomorphic function v = dw/dz = v
x
i v
y
) onto
A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301 1299
G
f
of Fig. 2b is carried out taking into account an additional vertex M of the cut CMB and with the correspondence of points
M? m. Again from the SchwarzChrsitoffel formula we write:
dw
dz
= a
2
_
f
0
s
b1=2
(c s)
c1=2
(s 1)
bc1
(ms) ds; (32)
where the single-valued branch of the integrand function f
b1/2
(c f)
c1/2
(f + 1)
b +c1
(m f) is xed in the upper half
plane by the condition of its positiveness at f = n, 0 < n < m.
The mapping parameter a
2
in (32), is found from the given Darcian velocity at point A i.e. V
A
= k
2
e
ip(1/2b)
sinpc/
sinp(b + c) that at f = 1 results in
a
2
=
k
2
sinpc
I
3
sinp(b c)
; I
3
=
_
1
0
t
b1=2
(c t)
c1=2
(1 t)
bc1
(t m) dt: (33)
After simple derivations
I
3
= c
1=2c
B(1=2 b; b c) m(1 1=c)
b1=2

1 2b
1 2c
F(3=2 b; 1=2 c; 3=2 c; 1=c)
_ _
:
In order to nd the last (and most difcult) mapping parameter m (see G
f
, Fig. 2b) we integrate (32) as
w(f) =
_
f
0
dw
dz
dz
df
df: (34)
In (34) dw/dz(f) is retrieved from (32) (taking into account (33)) and dz/df follows from (29) as
dz
df
=
L
I
1
f
b1
(c f)
c1
(1 f)
bc
: (35)
From (34) at f = c and w = k
2
H (see Fig. 2c) we obtain a non-linear equation with respect to m. We solve this equation by the
FindRoot routine using numerical integration by the NIntegrate routine of Mathematica.
Eventually, the ow rate is Q = w(1)/i. For nding Q we again implement the NIntegrate.
Appendix B
In this appendix we solve the refraction problem shown in Fig. 4a.
All streamlines in both S
1
and S
2
are straight. Correspondingly, recalling that we assumed h
AEO
= 0, from the Darcy law
applied to the segment EM and from the right-angled triangle OEM:
[
~
v
1
[ = k
1
0 h
M
[EM[
= k
1
h
M
[OM[ sinpb
; (36)
where h
M
and OM depend, of course, on the position of point M (but their ratio does not).
The refraction condition (5) is actually a geometrical combination of two general conditions (8). The second condition (8),
and (3) and (5) applied to EMN give:
[
~
v
1
[ cos pb = [
~
v
2
[ cos pb
1
: (37)
Next, we apply the Darcy law to the segment MN:
[
~
v
2
[ = k
2
h
M
h
N
[MN[
: (38)
In (38) h
N
= y
N
= [ON[ cospc. From the triangle OMN we have [ON[ = [OM[ cospb
1
/cosp(a + b
1
) and [MN[ = [OM[ sinpa/
cosp(a + b
1
) that upon substituting into (38), with provision (4), gives:
V
2

cos pb
1
cos pc
sinpa
=
h
M
cos p(a b
1
)
[OM[ sinpa
: (39)
Now, we exclude h
M
/[OM[ from (39) using (36) and get:
V
2

cos pb
1
cos pc
sinpa
= V
1
cos p(a b
1
) sinpb
sinpa
: (40)
Involvement of (37) leads to (4). The inequality (7) (overlooked by Kalinin [34]) follows from the condition that pressure
along BMO must be positive. Otherwise, the inltration shower regime (see PK77 and [35]) occurs in the downstream shell
and, hence, the analysis above is not valid.
1300 A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301
Appendix C. Supplementary data
Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.apm.2011.07.088.
References
[1] O.D.L. Strack, A.R. Kacimov, Application of mathematics to ow in porous media before the computer age; an introduction to the Special Issue
Applying mathematics to ow in porous media, J. Eng. Math. 64 (2009) 8184, doi:10.1007/s10665-009-9289-8.
[2] J. Crank, Free and Moving Boundary Problems, Clarendon Press, Oxford, 1984.
[3] J.D. Rice, J.M. Duncan, Deformation and cracking of seepage barriers in dams due to changes in the pore pressure regime, J. Geotech. Geoenviron. Eng.
136 (2010) 1625.
[4] Filter Diaphragms, National Engineering Handbook USDA, Natural Resources Conservation Service, Part 628, 2007.
[5] H.R. Cedergren, Seepage, Drainage, and Flow Nets, Wiley, New York, 1997.
[6] W.P. Creager, J.D. Justin, J. Hinds, Engineering for Dams V. 3. Earth, Rockll, Steel and Timber Dams pp. 619929, Wiley, New York, 1945.
[7] V.S. Istomina, Stability of Soils Against Seepage, Gosstrojizdat, Moscow, 1957 (in Russian).
[8] Gradation Design of Sand and Gravel Filters. National Engineering Handbook, USDA, Natural Resources Conservation Service Part 633, 1994.
[9] K.S. Richards, K.R. Reddy, Critical appraisal of piping phenomena in earth dams, Bull. Eng. Geol. Environ. 66 (2007) 381402, doi:10.1007/s10064-007-
0095-0.
[10] N.B. Ilyinsky, A.R. Kacimov, N.D. Yakimov, Analytical solutions of seepage theory problems. Inverse methods, variational theorems, optimization and
estimates (A review), Fluid Dyn. 33 (N2) (1998) 157168.
[11] R. Bea, D. Cobos-Roa, R. Storesund, Discussion of overview of New Orleans levee failures: lessons learned and their impact on national levee design and
assessment by G.L. Sills, N.D. Vroman, R.E. Wahl, N.T. Schwanz May 2008, D. J. Geotech. Eng. ASCE, 135, 19911994, vol. 134, No. 5, 2009, pp. 556565.
[12] N.N. Pavlovsky, On Seepage of Water Through Earth Dams, Kubuch, Leningrad, 1931 (in Russian).
[13] R. Dachler, Grundwasserstromung, Springer, Wien, 1936 (in German).
[14] B. Davison, L. Rosenhead, Some cases of the steady two-dimensional percolation of water through ground, Proc. R. Soc. A. 175 (962) (1940) 345365.
[15] A. Casagrande, Seepage through dams, in: Contributions to Soil Mechanics 19251940, Boston Society of Civil Engineers, Boston, MA, 1940, pp. 295
336.
[16] F.B. Nelson-Skornyakov, Seepage in Homogeneous Media, Sovetskaya Nauka, Moscow, 1949 (in Russian).
[17] V.I. Aravin, S.N. Numerov, Theory of Fluid Flow in Undeformable Porous Media, Gostekhizdat, Moscow, 1953. Engl. Translation by the Israel Program
Scient. Transl. Jerusalem, 1965.
[18] V.N. Aravin, S.N. Numerov, Calculation of Seepage in Hydraulic Structures, GILSA, Leningrad, 1955 (in Russian).
[19] P.Ya. Polubarinova-Kochina, Theory of Ground-Water Movement, Nauka, Moscow, 1977 (in Russian).
[20] R.V. Craster, Two related free boundary problems, IMA J. Appl. Math. 52 (1994) 253270.
[21] S.D. Howison, J.R. King, Explicit solutions to six free-boundary problems in uid ow and diffusion, IMA J. Appl. Math. 42 (1989) 155175.
[22] P.S. Huyakorn, G.F. Pinder, Computational Methods in Subsurface Flow, Academic Press, New York, 1983.
[23] A. Ouria, M. Mohammad, Application of NelderMead simplex method for unconned seepage problems, Appl. Math. Model. 33 (2009) 35893598.
[24] J. Bear, Looking forward. <http://timecapsule.ecodev.ch/video.html>, 2006.
[25] Yu.V. Obnosov, Periodic heterogeneous structures: new explicit solutions and effective characteristics of refraction of an imposed eld, SIAM J. Appl.
Math. 59 (1999) 12671287.
[26] Yu.V. Obnosov, A generalized MilneThomson theorem, Appl. Math. Lett. 19 (2006) 581586.
[27] Yu.V. Obnosov, A generalized MilneThomson theorem for the case of parabolic inclusion, Appl. Math. Model. 33 (2009) 19701981.
[28] Yu.V. Obnosov, Three-phase eccentric annulus subjected to a potential eld induced by arbitrary singularities, Quart. Appl. Math., 2011.
[29] A.R. Kacimov, Yu.V. Obnosov, Analytic solution to a problem of seepage in a checker-board massif, Transp. Porous Media 28 (1997) 109124.
[30] A.R. Kacimov, Yu.V. Obnosov, Steady water ow around parabolic cavities and parabolic inclusions in unsaturated and saturated soils, J. Hydrol. 238
(2000) 6577.
[31] A.R. Kacimov, Yu.V. Obnosov, N.D. Yakimov, Groundwater ow in a medium with a parquet-type conductivity distribution, J. Hydrol. 226 (1999) 242
249.
[32] Yu.V. Obnosov, R.G. Kasimova, A. Al-Maktoumi, A.R. Kacimov, Can heterogeneity of the near-wellbore rock cause extrema of the Darcian uid inow
rate from the formation (the Polubarinova-Kochina problem revisited)?, Comput Geosci. 36 (2010) 12521260, doi:10.1016/j.cageo.2010.01.014.
[33] B.B. Davison, Groundwater movement, in: N.E. Kochin (Ed.), Some New Problems of Mechanics of Continua, Akad. Nauk SSSR, Moscow, 1938, pp. 219
356 (in Russian).
[34] N.K. Kalinin, Seepage through a two-layered wedge, Prikl. Mat. Mekh. 16 (1952) 213222 (in Russian).
[35] A.R. Kacimov, Maximisation of water storage in back-lled and lined channels and dimples subject to evaporation and leakage, J. Irrig. Drain. Eng. 134
(2008) 101106.
[36] J.A. Liggett, P.L.F. Liu, Unsteady interzonal free surface ow in porous media, Water Resour. Res. 15 (1979) 240246.
[37] J.J. Rulon, R. Rodway, A. Freeze, The development of multiple seepage faces on layered slopes, Water Resour. Res. 21 (1985) 16251636.
[38] G.A. Fox, G.V. Wilson, The role of subsurface ow in hillslope and stream bank erosion: a review, Soil Sci. Soc. Amer. J. 74 (3) (2010) 713733.
[39] Y. Mualem, Interface refraction at the boundary between two porous media, Water Resour. Res. 9 (1973) 404414.
[40] P.A.C. Raats, Refraction of a uid at an interface between two anisotropic porous media, Z. Angew. Math. Phys. 24 (1973) 4353.
[41] G.A. Grinberg, The Selected Topics of the Mathematical Theory of the Electric and Magnetic Phenomena, Akad. Nauk SSSR, Moscow, 1948 (in Russian).
[42] Yu.V. Obnosov, The Solution of the problem of R-linear conjugation for the case of a hyperbolic boundary line of heterogeneous phases, Russian Math.
48 (2004) 5059.
[43] J. Bear, Dynamics of Fluids in Porous Media, Dover, New York, 1972.
[44] A.R. Kacimov, Yu.V. Obnosov, Semipermeable boundaries and heterogeneities: modeling by singularities, J. Hydrol. Eng. ASCE 6 (2001) 217224.
[45] A.R. Kacimov, Three-dimensional groundwater ow to a shallow pond: an explicit solution, J. Hydrol. 337 (2007) 200206.
[46] A.R. Kacimov, Yu.V. Obnosov, Analytical determination of seeping soil slopes of a constant exit gradient, Z. Angew. Math. Mech. 82 (2002) 363376.
[47] R.W. Goldstein, V.W. Entov, Qualitative Methods in Continuum Mechanics Longman, New York, 1994.
[48] O.D.L. Strack, Groundwater Mechanics, Prentice Hall, Englewood Cliffs, 1989.
[49] E.I. Anderson, Conformal mapping of groundwater ow elds with internal boundaries, Adv. Water Resour. 25 (2002) 279291.
[50] S. Johansson, P., Sjdahl, H. Cooney, Seepage Monitoring in Embankment dams using distributed temperature sensing in optical bres experience from
some Swedish dams, in: International Symposium on Modern Technology of Dams The 4th EADC Symposim, October 1218, Chengdu, China, 2007.
[51] S. Wolfram, Mathematica. A System for Doing Mathematics by Computer, Addison-Wesley, Redwood City, 1991.
[52] M. Abramowitz, I.A. Stegun, Handbook of Mathematical Functions, Dover, New York, 1970.
A. Kacimov, Y. Obnosov / Applied Mathematical Modelling 36 (2012) 12861301 1301