determination of groundwater flownets, fluxes, velocities
TRANSCRIPT
ARTICLE IN PRESS
0955-7997/$ - se
doi:10.1016/j.en
�CorrespondE-mail addr
Engineering Analysis with Boundary Elements 30 (2006) 1030–1044
www.elsevier.com/locate/enganabound
Determination of groundwater flownets, fluxes, velocities, and traveltimes using the complex variable boundary element method
Todd C. Rasmussena,�, Guoqing Yub
aWarnell School of Forestry and Natural Resources, University of Georgia, Athens, GA 30602-2152, USAbWater Resources and Hydroelectric Power Institute, Hohai University, Nanjing 210098, China
Received 11 August 2005; accepted 16 January 2006
Available online 7 November 2006
Abstract
The complex variable boundary element method, CVBEM, employs the Cauchy integral with any complex variable (e.g., complex
potential, complex flux, or complex velocity) to solve boundary value problems. The CVBEM formulation is consistent with the primal
and dual solutions of the boundary integral equation, as well as the analytic element method. The resulting problem is overdetermined
because two boundary conditions can be specified at each node. Ordinary least squares provides a unique solution that minimizes
boundary specification errors. Flownets are obtained by noting that the position of fluid–stream potential intersections can be found by
exchanging potential and position in the Cauchy integral, which enhances the determination of travel times along streamlines. Three
regional groundwater flow problems are used to illustrate the CVBEM approach, the original problem as defined by Toth, plus two
related problems as described by Domenico and Paliauskas, and by Nawalany.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Complex variable boundary integral method; Travel times; Groundwater flow and transport; Toth’s problem
1. Introduction
Mapping groundwater flow patterns is an important partof characterizing subsurface flow and transport behavior.A traditional tool for presenting maps of fluid potentialand streamlines is the flownet. The flownet helps tovisualize the flowfield, to delineate the capture zone orinfluence areas of boundary discharge and recharge, todesign groundwater development and remediation mea-sures, and to evaluate the effects of alternative boundaryconditions during site characterization.
Flownets are also used to compare calculated waterlevels with known observations while at the same timepredicting the path of contaminants along streamlines. Fortransport problems, flowpaths and travel times found byparticle tracking are used to identify outflow locations andthe arrival time of contaminants, as well as to show theadvective advance of a contaminant front within the flowdomain.
e front matter r 2006 Elsevier Ltd. All rights reserved.
ganabound.2006.01.017
ing author. Tel.: +1706 542 4300; fax: +1 706 542 8356.
ess: [email protected] (T.C. Rasmussen).
Extensive effort has been dedicated to the developmentof post-processing algorithms for mapping velocity fields.Among others, Pollock [1] used simple linear interpolationto generate the velocity vector field from a block-centered,finite-difference groundwater flow model. The Pollockapproach finds pathlines within a finite-difference grid cellusing an analytical function of nodal values. While thevelocity component is continuous in the direction of flow,discontinuities arise in the direction transverse to flow dueto the need to discretize the domain into grid cells.A continuous velocity field may also be obtained using
the finite-element method [2], which avoids the transversedirection discontinuity by using a bilinear interpolation forboth head and stream functions. Both finite-difference andfinite-element methods provide pathlines and travel times,and eliminate the need for time-stepping. The shortcomingof both techniques, however, is the need for internaldiscretization of the flow domain.Alternative methods for producing a groundwater
velocity field include mixed finite elements or mixed-hybridfinite elements [3,4]. These methods couple the headformulation with the velocity field by using head and
ARTICLE IN PRESST.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–1044 1031
velocity as degrees of freedom, or by applying the sameapproach to both head and velocity fields. Unfortunately,these methods suffer from the need to expend morethan twice as much computational effort as thestandard Galerkin models that use linear or bilinear headinterpolation.
Zijl [5,6] developed finite-element methods based on atransport velocity representation (TVR) of groundwatermotion. This method solves coupled Laplace-type equa-tions (two for two-dimensional flow fields, and three forthree-dimensional flow fields) and by adding auxiliaryboundary conditions to ensure that the problems arewell-posed.
While generally restricted to steady flow, these methodscan be applied to long-term contaminant transportproblems if short-term transients are unimportant andthe steady flow assumption is acceptable [7]. Nonetheless,these methods need to perform either a time-stepping orcontour interpolation when plotting flownets and calculat-ing travel times.
Regardless of the method employed, conventionalgroundwater flow models rely on either potential or streamfunctions as the primary unknowns, and determine fluidvelocities once these unknowns have been calculatednumerically. This dependence on potential or streamfunctions may result in poor velocity field accuracies whencompared to the accuracy of the primary unknowns.
Because the velocity field is usually deduced from theresults of flow modeling, the accuracy of the flow pathsdepends on both the flow model and the interpolationscheme used to construct the velocity field. Apart from theaccuracy of the velocity field, the accuracy of calculatedtravel times is also related to the step size associated withthe time-stepping scheme. Fine discretization in both spaceand time are major constraints to the application of bothflownet and pathline methods to travel time calculations.
A significant improvement in accuracy can be achievedby avoiding the discretization of the flow domain. Theboundary element method (BEM) is an established methodfor describing groundwater flow that reduces the problemdomain dimensionality by transforming a domain probleminto a boundary problem [8,9]. Thus, two-dimensionaldiscretization of the flow domain can be reduced to asimple discretization of the one-dimensional boundary,providing more accurate mapping of fluid potentials andstreamlines.
Another significant improvement was achieved byemploying the complex variable boundary element method(CVBEM), which provides a method for directly obtainingthe flownet [10,11]. CVBEM uses the Cauchy Integral toconvert a differential equation into a boundary integralequation which is further reduced to an algebraic systemafter approximation and discretization.
For many problems, CVBEM yields an overdeterminedsystem of linear algebraic equations that can be readilysolved using ordinary least squares (OLS). OLS incorpo-rates all of the original information, while at the same time
forming a square matrix that is symmetric, positive definiteand diagonally dominant [12,13].Another advance is the use of an inverse mapping
technique for defining the flownet, i.e., streamline andequipotential positions. Rather than solving for complexpotentials, yðzÞ ¼ fðzÞ þ {cðzÞ, as a function of complexposition (z ¼ xþ {y), we invert the map to provide thelocation of specified flownet intersections, zðyÞ ¼ xðyÞþ{yðyÞ.This flownet computation uses a post-processing strategy
that accurately and efficiently provides the flownetgeometry and travel times. Instead of determining traveltimes by summing travel-time increments between selectivelocations along a streamline, Dz, we obtain travel timesusing increments in fluid potential, Df, along eachstreamline.Yet another improvement is obtained by the direct
calculation of fluxes and flow velocities [30]. Two methodsare available; using the complex flux or velocity asboundary conditions, or by employing complex potentialboundary conditions and noting that the flux and velocityinternal to the domain is the derivative of the potentialfunctions. Both methods provide excellent predictions ofthe flux and velocity fields within the flow domain.
2. Mathematical formulation
2.1. Vector representation
The two-dimensional, steady continuity equation forhomogeneous, isotropic porous medium flow withoutinternal sources or sinks, and with negligible compressi-bility of the fluid and medium, can be written in terms ofthe divergence of the flux vector, ~q, using
r �~q ¼qqx
qxþ
qqy
qy¼ 0, (1)
where
~q ¼ ½qx; qy� ¼ rf (2)
is the Darcian flux vector, qx and qy are the compo-nents of flux in x- and y-directions, respectively,and f is the Grinskii, or discharge, potential [14], definedusing
f ¼
�Kh general case;
�Th confined aquifers;
�12
Kh2 unconfined aquifers;
8><>: (3)
where h is the hydraulic head, K is the homogeneous,isotropic hydraulic conductivity, T is the aquifer transmis-sivity, and the Dupuit–Forchheimer assumptions arerequired for the unconfined case.Eq. (1) is written in terms of the discharge potential using
r2f ¼q2fqx2þ
q2fqy2¼ 0. (4)
ARTICLE IN PRESST.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–10441032
Eq. (1) holds throughout the flow domain for both steadyflow and unsteady, incompressible flow. The curl of the fluxvector for these conditions is
r �~q ¼ det
qqx
qqy
qx qy
24
35 ¼ qqy
qx�
qqx
qy¼ 0 (5)
which is the same as
r � rf ¼qðqf=qyÞ
qx�
qðqf=qxÞ
qy¼ 0. (6)
The fluid velocity, ~v, is related to the flux using ~v ¼ ~q=Z,where Z is the effective porosity of the medium.A velocity potential, j ¼ f=Z, can also be formed ina manner similar to the discharge potential for homo-geneous conditions:
~v ¼ rj ¼ rfZ
� �(7)
for the same conditions as Eq. (3).
2.2. Complex representation
The vector representation can be written in complexform using
q ¼ qx þ {qy. (8)
The derivative of the complex flux is
rq ¼dq
dz¼
qx¼
qqx
qx� {
qqy
qx, (9)
where q ¼ qx � {qy is the complex conjugate of q, and z ¼
xþ {y is the complex position. The derivative can also befound using
rq ¼qq
qð{yÞ¼ �
qqy
qy� {
qqx
qy(10)
because the path of the derivative is immaterial. SettingEqs. (9) and (10) equal yields
rq ¼qqx
qx� {
qqy
qx¼ �
qqy
qy� {
qqx
qy(11)
which is the same as
qqx
qxþ
qqy
qy
� �� {
qqy
qx�
qqx
qy
� �¼ ðr �~qÞ � {ðr �~qÞ ¼ 0
(12)
for steady, irrotational flow.The complex potential for steady, homogeneous flow can
be formed using
y ¼ fþ {c, (13)
where f and c are the fluid potential and streamfunctions, respectively. The derivative of the complexpotential is
dydz¼
qyqx¼
qfqxþ {
qcqx
(14)
which is the same as
dydz¼
qyq ð{yÞ
¼qcqy� {
qfqy
. (15)
Setting Eqs. (14) and (15) equal yields
dydz¼
qfqxþ {
qcqx¼
qcqy� {
qfqy
(16)
which yields the Cauchy–Riemann conditions:
qx ¼qfqx¼
qcqy
,
qy ¼qfqy¼ �
qcqx
ð17Þ
so that q ¼ dy=dz and r2y ¼ 0. While f in Eq. (3) issuitable for general analysis of groundwater flow, thisanalysis shows that one must utilize the complex discharge,q, instead of complex potential, f, for both confined andunconfined aquifers in order for the Cauchy–Riemannconditions to hold.
2.3. Cauchy integral equation
The Cauchy theorem states that if a function, f ðzÞ, isanalytic in a single connected region, O, then the integral off ðzÞ with respect to z along a simple closed contour, G 2 O,must vanish:I
Gf ðzÞdz ¼ 0. (18)
The analytic function can be defined using
f ðzÞ ¼yðzÞ
z� a. (19)
Alternatively, the function can be defined using theconjugate of the complex flux:
f ðzÞ ¼qðzÞ
z� a. (20)
Yet another definition uses the complex velocity,v ¼ ðqx þ {qyÞ=Z, where Z is the effective porosity of themedium:
f ðzÞ ¼vðzÞ
z� a. (21)
The salient feature of the alternative formulations is thedecomposition of the analytic function into a pair oforthogonal components.Point a in Eqs. (19)–(21) can be located either inside or
along the boundary of the domain. For the case wherePoint a is located on the boundary, G:
oðaÞ ¼1
b{
IG
oðzÞz� a
dz; a 2 G (22)
and for Point a located within the domain, O:
oðaÞ ¼1
2p{
IG
oðzÞz� a
dz; a 2 O (23)
ARTICLE IN PRESST.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–1044 1033
where b is the inner angle at Point a made by tangentiallines before and after the point (b ¼ p for a smoothboundary at the point) and o can be taken as the complexpotential, y, complex flux, q ¼ dy=dz, or complex velocity,v ¼ q=Z.
2.4. CVBEM vs. BEM and AEM
A direct relationship exists between the traditional BEMand CVBEM. BEM uses Green’s second identity to find thevalue of potential within the flow domain [15,9]. We startby noting that
dz
z� a¼ d lnðz� aÞ ¼ d ln rþ {dl, (24)
where z� a ¼ re{l and a single-valued branch of ðp;�p� istaken for the argument of the complex logarithmicfunction. This means that Eq. (23), when combined withEq. (19), can be written as
yðaÞ ¼1
2p{
IGðfþ {cÞ � ðd ln rþ {dlÞ
¼1
2p{
IG
fd ln r� cqlqs
ds
� ��
þ{ cd ln rþ fqlqs
ds
� ��, ð25Þ
where n and s are the outwardly directed normal and thecounterclockwise oriented tangent vectors to the boundary,respectively. Noting that q ln r=qn ¼ ql=qs and performingan integration by parts yields
yðaÞ ¼1
2p
IG
fq ln r
qn� ln r
qcqs
� �ds,
þ{
2p
IG
cq ln r
qnþ ln r
qfqs
� �ds ð26Þ
which can be separated to yield
fðaÞ ¼1
2p
IG
fq ln r
qn� ln r
qcqs
� �ds,
cðaÞ ¼1
2p
IG
cq ln r
qnþ ln r
qfqs
� �ds. ð27Þ
These equations indicate that the real and imaginarycomponents of potential at Point a can be calculatedfrom the sum of the double-layer logarithmic potential(with densities f=2p, c=2p) and the single-layerlogarithmic potential (with densities ðqc=qsÞ=2p,�ðqf=qsÞ=2p) over the contour G. This assumesthat the complex potential along the inside of the bou-ndary equals the jump across the boundary, sothat the complex potential vanishes outside theboundary.
Finally, we obtain the conventional real-value boundaryelement equations by using the Cauchy–Riemann
conditions, qf=qn ¼ qc=qs and qf=qs ¼ �qc=qn:
fðaÞ ¼1
2p
IG
fq ln r
qn� ln r
qfqn
� �ds,
cðaÞ ¼1
2p
IG
cq ln r
qn� ln r
qcqn
� �ds ð28Þ
which are the primal solution that solves the divergenceproblem, r �~q ¼ 0, and the dual solution for the curlproblem, r �~q ¼ 0, respectively [7]. Note that thisformulation can also be used to find the complex flux,qðzÞ, and velocity, vðzÞ, using appropriate boundaryconditions.The analytic element method (AEM) can also be
compared to the CVBEM. AEM uses the superpositionprinciple to combine analytic solutions for individualpoints, lines, planes, etc. to construct the response withina domain to the combination of these geometries [16–18].Both the BEM and CVBEM normally place the modeled
domain within the boundary. Yet BEM can also be usedwhen the modeled domain is external to the boundary [9].For CVBEM, we can move a point source withinan infinite domain to the inside of the domain by insertinga cutting line that starts at the point source, extendsto infinity, and surrounds the domain. The distanceto the boundary is infinite except for the point source.This yields:
yðzÞ ¼ �Q
2p
IGdðz0Þd ln z, (29)
where Q is the total flow from the point source and dðz0Þ isthe Dirac delta function. This equation is equivalent to theAEM solution for a point source:
yðzÞ ¼ y0 �Q
2pln z=z0. (30)
3. Numerical implementation
3.1. Boundary value specification
Using the Cauchy integral equation to solve a boundaryvalue problem requires the specification of the boundaryconditions, wðaÞ where a 2 G. AEM is appropriate whenthese boundary conditions can be defined analytically. Forapplications where simple boundary conditions are notavailable, then the flow boundary is discretized in spaceand then constant or interpolated functions are used toapproximate the boundary conditions along each bound-ary segment.In this study, the boundary is approximated using
straight-line segments joined at their ends. Twokinds of boundary-value distributions along the elementare considered—constant and linearly varying boundaryvalues.
ARTICLE IN PRESST.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–10441034
For constant boundary values along each element,Eq. (22), becomes
oðziÞ ¼1
p{
Xn
j¼1jai
oj lnzjþ1 � zi
zj � zi
(31)
because b ¼ p for straight-line segments with constantboundary values when zi is defined at the centers ofsegments. Also, i ¼ 1; 2; . . . ; n, o is the complex function atzi, oj is the constant complex function at note zj in elementj, zj and zjþ1 are the complex positions of the endpoints ofelement j, and n is the total number of the boundaryelements.
In a similar manner, the solution within the domain,Eq. (23), becomes
oðziÞ ¼1
2p{
Xn
j¼1
oj lnzjþ1 � zi
zj � zi
. (32)
Separating the real and imaginary parts from Eq. (31), oneobtains 2n equations expressed in terms of nodal values ofo as
RðoðziÞÞ ¼1
p
Xn
j¼1jai
½�bijRðojÞ þ aijIðojÞ�,
IðoðziÞÞ ¼ �1
p
Xn
j¼1jai
½aijRðojÞ þ bijIðojÞ� ð33Þ
in which
aij ¼ lnzjþ1 � zi
zj � zi
��������,
bij ¼ � argzjþ1 � zi
zj � zi
� �, ð34Þ
where aii ¼ �P
iaj aij, bii ¼ �P
iaj bij. This can be writtenin matrix form using
A B
�B A
� �ð2n� 2nÞ
RðoÞ
IðoÞ
" #
ð2n� 1Þ
¼
0
0
� �ð2n� 1Þ
. (35)
The complex function for a linear interpolation ofboundary conditions is
oðzÞ ¼ N1ðzÞoj þN2ðzÞojþ1 (36)
for z 2 ½zj ; zjþ1�, and where
N1ðzÞ ¼zjþ1 � z
zjþ1 � zj
,
N2ðzÞ ¼z� zj
zjþ1 � zj
.
After insertion of Eq. (36) into Eq. (22) and integration,we obtain
oðziÞ ¼1
{bi � ln jðziþ1 � ziÞ=ðzi�1 � ziÞj
�Xn
j¼1jai
jai�1
½N1ðziÞoj þN2ðziÞojþ1� lnzjþ1 � zi
zj � zi
� �, ð37Þ
where i ¼ 1; 2; . . . ; n, and the other notation has the samemeaning as before. Eq. (37) can be rewritten as
Xn
j¼1
½aij � { bij�oj ¼ 0, (38)
where i ¼ 1; 2; . . . ; n and
aij ¼ RXGe3zj
ZGe
NjðzÞ
z� zj
dz
24
35,
bij ¼ � IXGe3zj
ZGe
NjðzÞ
z� zj
dz
24
35þ dijbi
in which dij is the Kronecker delta and the integration isperformed for the Cauchy principal value when i ¼ j. Thesingle-valued branch of the argument of complex logarith-mic function is taken in the same manner as in the constantdistribution case. Separating the real and imaginary partsyields a set of equations which are identical to Eq. (35) inshape, but have different matrix entries because of thedifference in interpolation functions.
3.2. Ordinary least squares
It can be observed that Eq. (35) results in a system of 2n
equations, regardless of which kind of boundary valueapproximation method we use. In addition, both methodscontain 2n nodal variables, with n variables for the realpart and n variables for the imaginary part. Theseequations usually require that a single boundary condition(e.g., either fluid potential, f, or stream function, c) beprescribed at every boundary node.By moving the data with known boundary conditions
(and their coefficients) to the right-hand side of theequation, and retaining the unknown nodal variables andcoefficients on the left-hand side, we obtain
Cu
RðoÞ
IðoÞ
" #u
¼ �Ck
RðoÞ
IðoÞ
" #k
, (39)
where the subscripts u and k refer to unknown and knownboundary nodal data, respectively. The unknown matrixhas the dimension ð2n�mÞ, where mo2n.The system of linear equations presented as Eq. (39) is
algebraically overdetermined; i.e., there are more equationsthan unknowns. This results from the specification of either
ARTICLE IN PRESST.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–1044 1035
the real or imaginary components of the flux at each node,thus reducing the number of unknowns.
For a well-posed problem containing known boundaryconditions on m boundary segments, there will generally be2m choices of the equation combinations and thus 2m
alternative solutions, each of which satisfies a subset of theoriginal equations.
The algebraic system of linear equations is commonlysolved using explicit, implicit or hybrid methods [11,19,20].The explicit and implicit methods selectively excluderelevant equations until the matrix is square, resulting indifferent approximate solutions depending upon whichequations were excluded. Although the hybrid methodprovides a minor improvement in solution accuracy, themethod requires a substantial increase in computationalload.
OLS is useful for solving overdetermined systems of[12,21]. The goal is to minimize the global approximationerror; i.e., the L2 norm of the residual error vector isminimized when the solution vector is substituted into thesystem of linear equations. Jankovic and Barnes [22] use asimilar approach to fit higher-order line elements toimprove the accuracy.
Employing the OLS method leads to [12]:
ou ¼ �½CTu Cu�
�1½CTu Ck�ok, (40)
where the superscript, T, refers to the transpose of theoriginal matrix, and the exponent, �1, refers to the inverseof the matrix. Eq. (40) is our working equation system forthe solution vector. The coefficient matrix of Eq. (40) issymmetric, positive definite and diagonally dominant,which simplifies its solution [12,13].
3.3. Flownet mapping
Defining the flownet requires finding the position ofstreamline–equipotential intersections. Our approach is totransform the complex potential problem in the physicalplane, yðzÞ, into a complex position problem in thepotential plane, zðyÞ. Note that the differentiation rule ona compound function f ½fðx; yÞ;cðx; yÞ�, leads to
qfqx
qfqx
24
35 ¼ qf
qxqcqx
qfqy
qcqy
24
35 qf
qf
qfqc
24
35 (41)
or
qfqf
qfqc
24
35 ¼ 1
Jb
qcqy
�qcqx
�qfqy
qfqx
24
35 qf
qx
qfqy
24
35, (42)
where Jb is the Jacobian:
Jb ¼ det
qfqx
qcqx
qfqy
qcqy
24
35 ¼ q2
x þ q2y ¼ kqk
2. (43)
By applying the differentiation rule to x and y, we obtain
qxqf
qxqc
24
35 ¼ 1
kqk2
qfqy
�qcqy
264
375,
qyqf
qyqc
264
375 ¼ 1
kqk2
�qcqx
qfqx
24
35. ð44Þ
If the fluid flux can be determined (i.e., finite and nonzero),we have
qx
qf¼
qy
qc¼
qx
kqk2,
qy
qf¼ �
qx
qc¼
qy
kqk2ð45Þ
and
q2x
qf2þ
q2x
qc2¼ 0,
q2y
qf2þ
q2y
qc2¼ 0 ð46Þ
or, in complex notation:
q2z
qf2þ
q2z
qc2¼ 0. (47)
Eq. (47) implies that the position of the complexpotential can be defined in the potential plane, and isthe starting point for the inverse mapping procedure.This equation can be used to solve for the position offlownet intersections within the flow domain once all ofthe complex potential boundary conditions have beendetermined,As noted above, the complex potential at any point
within the flow domain is found using
oðzÞ ¼1
2p{
Xn
k¼1
ok
zkþ1 � z
zkþ1 � zk
� ��
�okþ1zk � z
zkþ1 � zk
� ��ln
zkþ1 � z
zk � z
� �. ð48Þ
From Eq. (47), it is clear that the position of a target fluidpotential–stream function intersection, zðytÞ, can be foundby term-by-term swapping of the position with thepotential in the Cauchy integral:
zðytÞ ¼1
2p{
IG
zðyÞy� yt
do (49)
which is identical to Eq. (23) except that thecomplex potential and position have been exchanged.Upon discretization with linear interpolation along linear
ARTICLE IN PRESST.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–10441036
boundary elements, this becomes
zðoÞ ¼1
2p{
Xn
k¼1
zk
okþ1 � ookþ1 � ok
� ��
�zkþ1ok � o
okþ1 � ok
� ��ln
okþ1 � ook � o
� �. ð50Þ
For cases when the transformed domains in the complexpotential plane include a branch cut (i.e., complex positionsoverlap in the complex potential domain), a prediction–correction procedure can be employed to locate branchcuts (i.e., ground-water divides) that isolate regions of one-to-one correspondence. The procedure can be applied ineither an upstream or downstream direction. This ap-proach is especially useful when trying to determine theextent of capture zones [23].
Note that the inverse mapping procedure can be used forany of the functions specified in Eqs. (19)–(21), i.e.,o ¼ ½y; q; v�.
3.4. Flux and velocity
We present two approaches for solving flux and velocityproblems. The first is to ignore the potential problem andrestrict our solution to flux or velocity, only. To do this wedirectly solve the Cauchy Integral explicitly for flux orvelocity using Eq. (20) or (21), respectively, with theircorresponding boundary conditions.
qðaÞ ¼1
2p{
IG
q
z� adz a 2 O, (51)
vðaÞ ¼1
2p{
IG
v
z� adz a 2 O. (52)
Eqs. (51) and (52) provide direct estimates of the flux andvelocity at any point within the domain, O. Thisapproach is suitable when the fluid potential is notrelevant to the problem, such as when all of theboundary conditions are specified in terms of flux orvelocity. This method takes advantage of the fact that both
Fig. 1. Complex flux problem geometry. The flux vector, q ¼ qx þ {qy, can
be separated into normal, qn, and tangential, qs, components along the
boundary, so that q ¼ qn þ {qs.
flux and velocity can be separated into orthogonalcomponents, allowing them to serve as the primaryvariable in the Cauchy integral.The boundary conditions for this formulation are
specified using the normal and tangential components offlux, q ¼ qn þ {qs, or velocity, v ¼ vn þ {vs, shown in Fig. 1.The boundary normal vector, ~n, is pointed outward fromthe boundary, and the boundary tangential vector, ~s ¼ {~n,is pointed counterclockwise along the boundary, keepingthe domain to the left of the boundary.The use of conventional boundary conditions—i.e., the
specified potential (Dirichlet) and specified flux(Neumann)—implies a specification of the boundaryflux. That is, the first type implies the specificationof the tangential component of the complex flux, whilethe second type implies the specification of the normalcomponent of the complex flux [23]. For example, qs ¼ 0results from the specification of a constant potential alonga boundary, qf=qs ¼ 0, and qn ¼ 0 results from thespecification of an impermeable boundary condition,qf=qn ¼ 0.A second method for determining the flux and velocity
within the flow domain relies on the fact that thecomplex potential is continuously differentiablewithin the flow domain, so that a continuous velocityfield can be defined within the flow domain usingcomplex boundary potentials [24]. Eq. (23) has alreadybeen used to define the complex potential within the flowdomain:
yðaÞ ¼1
2p{
IG
yz� a
dz; a 2 O. (53)
The flux is the derivative of Eq. (53) with respect toposition:
qðaÞ ¼dydz
����a
¼1
2p{
IG
y
ðz� aÞ2dz; a 2 O (54)
which can be approximated using
q ¼1
2p{
Xn
k¼1
gkyk, (55)
where linear interpolation of y along G is used, and:
gk ¼1
zk � zk�1
� �ln
zk � z
zk�1 � z
� �
�1
zkþ1 � zk
� �ln
zkþ1 � z
zk � z
� �. ð56Þ
The fluid velocity at any point can be found by dividing theflux by the effective porosity, v ¼ q=Z. The resulting fluxand velocity fields are continuous because (a) the right-hand side terms of Eq. (56) include elementary functionsonly, (b) the points zj ; zjþ1; etc., are located on theboundary, and (c) the point zi is always located inside thedomain.Eq. (55) can be used once all the complex boundary
conditions have been specified. If some of the boundary
ARTICLE IN PRESST.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–1044 1037
conditions are unknown, then they must first be obtainedbefore Eq. (55) can be used to determine the complex fluxat locations internal to the flow domain. It is clear thatspecification of the flow problem using Eq. (55) results in aunique analytic solution of the complex flux and velocitythroughout the flow domain.
3.5. Travel times
The travel time, t, of a fluid particle moving from Point a
to Point b along a streamline is the path integral of theinverse velocity [13]:
t ¼Z b
a
ds
v, (57)
where s is the distance along the streamline and v ¼
vx þ { vy ¼ q=Z is the fluid velocity, and where q is the fluidflux and Z is the effective porosity. Also note that q ¼ rywhere y ¼ fþ {c, so that
t ¼1
Z
Z fb
fa
dfkvk2
(58)
because qc=qs ¼ 0 along a streamline, and kvk2 ¼ v v.We can also write the travel time as
t ¼ ZZ fb
fa
kuk2 df, (59)
where
u ¼dz
dy¼
qx
qfþ {
qy
qf¼
q
Jb
¼1
q. (60)
We now have an integral equation in the complex potentialplane that provides the travel time of a particle along astreamline between two potentials. The integral can readilybe solved by first transforming the interval from naturalvalues, f 2 ½fa;fb� to x 2 ½�1; 1�:
f ¼1� x2
fa þ1þ x2
fb (61)
so that
t ¼ ZDf2
Z 1
�1
kuk2 dx, (62)
where Df ¼ fb � fa. Quadrature leads to
t ¼ ZDf2
Xj
Gjkuk2j , (63)
where j indicates the jth quadrature point, Gj and kuk2j are
the weighting factor and integrand values at point xj. Thevalues of c and Df are normally prescribed while thevalues of u are calculated using the complex potentialanalog to the complex position derivative:
uðyÞ ¼1
2p{
Xn
k¼1
f kzk, (64)
where
f k ¼ln½ðyk � yÞ=ðyk�1 � yÞ�
yk � yk�1�
ln½ðykþ1 � yÞ=ðyk � yÞ�ykþ1 � yk
.
(65)
4. Applications
Three analytic solutions are used to demonstrate thestrengths and weaknesses of the modeling approachpresented here. These solutions are variations of a solutionpresented by Toth in 1963 [25], which describes regionalgroundwater flow in a rectangular domain. The solution isappropriate for steady, two-dimensional (profile), satu-rated, homogeneous, isotropic flow bounded above by awater table and bounded on the remaining three sides byimpermeable surfaces.The Toth solution is appropriate for a linear trend
plus a sinusoidal variation in water-table elevation.Two additional solutions for this problem are alsoavailable. Domenico and Palciauskas [26] and Nawalany[27] provide additional numerical solutions for slightlydifferent water-table boundary conditions. These solutionsare used to provide the complex potentials and fluxeswithin the flow domain for comparison with the CVBEMsolution.The problem geometry consists of a domain two
horizontal units in distance and a unit thickness. A totalof 60 nodes at equal spacing are distributed alongthe domain boundary. The problem geometry is shown inFig. 2. A representative water-table boundary condition isalso shown. Note that flow between the water-table and therectangular flow domain is not considered.
4.1. Toth’s problem
In 1963, Toth [25] presented a solution for regional, two-dimensional, ðx; yÞ, groundwater flow in an unconfined,homogeneous, and isotropic aquifer. A symmetrical land-scape is assumed, with no-flow boundaries on the verticalsides and horizontal bottom of the domain. x0 and y0 arethe horizontal and vertical extent of the flow domain,respectively. The upper, horizontal surface is specifiedusing
fðx; yÞ ¼ y0 þ cxþ a sin bx, (66)
where c ¼ tan a, a ¼ a0= cos a, and b ¼ b0= cos a, and wherea, is the slope of the regional water table perpendicular tothe valley bottom, and a0 and b0 are the amplitude andfrequency of water-table fluctuations about the regionalslope.Toth’s analytic solution for this problem is [25]
fðx; yÞ ¼ y0 þcx0
2þ
a
bx0ð1� cos bx0Þ
þX1j¼1
b cos ax cosh ay, ð67Þ
ARTICLE IN PRESS
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1 41
11 31
Distance from flood plain
Ele
vatio
n
Fig. 2. Problem domain geometry. Horizontal extent is two units. Vertical extent is one unit. Nodal locations indicated using open circles (�). Sides and
lower boundaries are impermeable. Top boundary condition is specified using a function that changes for each of the three examples (the Toth problem
boundary condition is illustrated).
Potential Contours Analytic
Flux Contours Analytic
(a)
(b)
Fig. 3. Toth analytic solutions for (a) complex potential, y ¼ fþ {c and
(b) complex flux, q ¼ qx þ {qy. Solid and dotted lines show constant values
for the real and imaginary components, respectively.
T.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–10441038
where a ¼ jp=x0 and
b ¼2
x0 cosh ay0
abð1� cos bx0 cos jpÞ
b2� a2
�
þcðcos jp� 1Þ
a2
�. ð68Þ
The flux vectors are obtained by differentiating withrespect to x and y, respectively:
qx ¼qfqx¼ �
X1j¼1
ab sin ax cosh ay, (69)
qy ¼qfqy¼X1j¼1
ab cos ax sinh ay (70)
and the stream function is readily found to be
cðx; yÞ ¼ c0 �X1j¼1
b sin ax sinh ay (71)
which was obtained using the Cauchy–Riemann condi-tions.
A linear interpolator is used for boundary conditionsbetween the 60 nodes. The first simulation uses prescribedhead boundary conditions, fðxbÞ, on the upper surface andprescribed stream function boundary conditions, c ¼ 0, onthe sides and bottom. The second simulation uses prescribedtangential flux, qsðxbÞ, along the upper boundary, andprescribed normal flux, qn ¼ 0, along the sides and bottom.
ARTICLE IN PRESS
Potential Contours Numeric Potential Error
0.00
02
0.0002
0.0002
Flux Potential Boundary Conditions Flux Error
0.0002
0.0002
0.00020. 0005
0 .00 0 5
0.0005
0 .00050.00 1
0.00
1
0 .001
0 001
0.0
02
0.0
02
0.005
Flux Flux Boundary Conditions Flux Error0.0005
0.001
0.001
0.002
0.002
(a)
(b)
(c) (f )
(e)
(d)
Fig. 4. Toth numeric solutions for (a) complex potential, y ¼ fþ {c, (b) complex flux, q ¼ qx þ {qy, calculated using boundary potentials, yb,
and (c) complex flux calculated using boundary fluxes, qb. Contours of error norms between the analytic and numeric solutions are shown in (d)–(f).
1 1.1 1.2 1.3 1.4 1.50.02
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
φ
ψ
Position Contours Numeric
1 1 1 3 1 41
42
44
46
48
50
5254
56
58
60
Fig. 5. Contour lines of constant complex position, z ¼ xþ {y, within the complex potential domain, y ¼ fþ {c. Lines of constant horizontal distance, x,
and elevation, y, are shown using solid and dotted lines, respectively. Flownet positions can be found using constant increments in complex potential,
Df and Dc, respectively.
T.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–1044 1039
ARTICLE IN PRESST.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–10441040
Figs. 3 and 4 present the analytic and calculatedcomplex potentials and fluxes for this problem. Fig. 4aillustrates the resulting flownet, which are just lines ofequal complex potential. Fig. 4d shows the magnitude ofthe errors, e ¼ jy� yj, between the analytic and numericsolutions.
To illustrate how the complex position of equi-potential–streamline intersections are obtained, a plot ofthe complex positions as a function of complex po-tential calculated using Eq. (50) is presented as in Fig. 5.Lines of constant horizontal and vertical extent are shownusing the solid and dotted lines, respectively. Travel timescan be obtained using Eq. (63) along each streamline,c ¼ ci.
Horizontal (solid lines) and vertical (dashed lines) fluxcontours are presented for prescribed complex potentialboundary conditions in Fig. 4b. Fig. 4e presents themagnitude of the calculated flux error, e ¼ jq� qj. Notethat the largest errors occur near the boundary of the flowdomain.
Calculated flux contours obtained using complex fluxboundary conditions are shown in Fig. 4c, and thecorresponding errors are shown in Fig. 4f. Note theimproved estimate in flux (i.e., smaller errors) nearthe boundary. This reduction in overall error near the
Potential Cont
11
Flux Contou
11
1 4
1 4
(a)
(b)
Fig. 6. Domenico and Palciauskas analytic solutions for (a) complex potential,
constant values for the real and imaginary components, respectively.
boundary is offset by slightly poorer velocity estimateswithin the interior of the domain.
4.2. Domenico and Palciauskas solution
Domenico and Palciauskas [26] present a simplifiedboundary condition that uses a cosine variation ofhydraulic head prescribed along the top boundary:
f ¼ cospx
2(72)
and no flow on the remainder of the boundaries, and wherex0 ¼ 2. These conditions are equivalent to:
our
rs
y ¼
Top boundary. Nodes 41 to 1: qx ¼ �p=2 sinðpx=2Þ.Left boundary. Nodes 1 to 11: qx ¼ 0.Lower boundary. Nodes 11 to 31: qy ¼ 0.Right boundary. Nodes 31 to 41: qx ¼ 0
where a unit hydraulic conductivity is assumed. Theanalytic solution for this problem is [26]
qx ¼ �p2sin
px
2cosh
py
2� tanh
p2
sinhpy
2
h i,
qy ¼p2cos
px
2sinh
py
2� tanh
p2
coshpy
2
h ið73Þ
s Analytic
31
Analytic
31
1
1
fþ {c, and (b) complex flux, q ¼ qx þ {qy. Solid and dotted lines show
ARTICLE IN PRESS
Potential Contours Numeric Potential Error
0.0002
0.0004
0.0006
0.0008
0.0008
0.0010.0010.0012
0.0014
Flux Potential Boundary Conditions Flux Error
0.0005
0.0005
0.001
0.0010.001
0.0010.001 5
0.0015 0.0015
0 .00150.0 02
0.002 0.002
0.00 20.0 03
0.0030.003
0.003
Flux Flux Boundary Conditions Flux Error
0.001
0.00
1
0 .0 0 1 5 0.0015
0.0020.0020.0025
(a)
(b)
(c)
(d)
(e)
(f)
Fig. 7. Domenico and Palciauskas numeric solutions for (a) complex potential, y ¼ fþ {c, (b) complex flux, q ¼ qx þ {qy, calculated using boundary
potentials, yb, and (c) complex flux calculated using boundary fluxes, qb. Contours of error norms between the analytic and numeric solutions are shown in
(d)–(f).
T.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–1044 1041
and the fluid potential and stream functions are
f ¼ cospx
2cosh
py
2� tanh
p2
sinhpy
2
h i,
c ¼ � sinpx
2sinh
py
2� tanh
p2
coshpy
2
h i. ð74Þ
The analytic solution for both the complex potential andflux is shown in Fig. 6. The numerical solution is obtainedby again using linear interpolation between boundarynodes. The results, shown in Fig. 7, are virtually identicalto the analytic solution. Note that the velocity field errorsare practically identical, but the errors near the boundaryare again greater when complex potentials are used as theboundary conditions.
Solutions using the finite-element method based on aTVR, Zijl [5] used 40 and 20 equidistant intervals in the x-and y-directions, respectively, for the discretization of thedomain and obtained similar results. The clear advantageof the boundary integral method lies in the simplicity ofdiscretization, which is limited to a single dimension alongthe boundary.
4.3. Nawalany solution
Another simplification of the Toth problem was devel-oped by Nawalany [27]. In this problem a sinusoidal head
is superimposed on a linear trend along the topboundary:
f ¼ �c x�x0
2psin
2px
x0
� �(75)
and the rest of the boundary conditions are identical. Thisis equivalent to:
Top boundary. Nodes 41 to 1: qx ¼ �cð1� cos 2px=x0Þ.Left boundary. Nodes 1 to 11: qx ¼ 0.Lower boundary. Nodes 11 to 31: qy ¼ 0.Right boundary. Nodes 31 to 41: qx ¼ 0.The analytic solution for this problem is
qx ¼X1m¼1
ab sin ðaxÞ cosh ð�aðy0 � yÞÞ, (76)
qy ¼ �X1m¼1
ab cos ðaxÞ sinh ð�aðy0 � yÞÞ, (77)
where a ¼ mp=x0, and
b ¼8cx0ð1� cos mpÞ
ðmpÞ2ðm2 � 4Þ coshð�ay0Þ(78)
when m is odd and b ¼ 0 when m is even.
ARTICLE IN PRESS
Potential Contours Analytic
11 31
Flux Contours Analytic
1 4 1
1 4 1
11 31
(a)
(b)
Fig. 8. Nawalany analytic solutions for (a) complex potential, y ¼ fþ {c,and (b) complex flux, q ¼ qx þ {qy. Solid and dotted lines show constant
values for the real and imaginary components, respectively.
T.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–10441042
The equipotentials and streamlines can be obtainedusing integration:
f ¼ f0 �X1m¼1
b cos ðaxÞ cosh ð�aðy0 � yÞÞ, (79)
c ¼ c0 þX1m¼1
b sin ðaxÞ sinh ð�aðy0 � yÞÞ. (80)
For our problem, we set c ¼ 0:1, x0 ¼ 1, and y0 ¼ 1, andagain use 60 elements along the perimeter. Results areshown in Figs. 8 and 9 for the analytic and simulationresults. Note that the results are consistent with both theToth and Nawalany solutions.
Table 1 provides a comparison between the methodpresented here with the analytic solution [26], TVR, andconventional FEM solutions using 88 elements [27]. It isclear that our method is superior to both the FEM andTVR methods, and requires less discretization effort.
5. Discussion and conclusions
Internal discretization of the flow domain can be avoidedfor conditions of steady flow through homogeneous mediausing real-variable BEMs, complex-variable boundaryelement methods (CVBEMs), and analytic element meth-ods (AEMs). The advantage of AEM and CVBEM lies in
their ability to determine both streamlines and velocities atpoints within the flow domain with excellent accuracy andcontinuity. This capability lays the foundation upon whichcontaminant transport problems can be addressed.CVBEM uses the Cauchy integral along with boundary
elements to convert a domain problem into a boundaryproblem, which simplifies the solution of domain-typeproblems by reducing the discretization and computationaleffort. CVBEM is more efficient and accurate than BEMbecause of the reduced number of equations requiringsolution.By reformulating the problem in the complex potential
plane, the same code can be used to compute the complexpositions of specified complex potentials for appropriateboundary conditions. This provides us with the positions ofthe flownet (equipotential–streamline) intersections withgreater accuracy and efficiency.Travel times can be accurately determined using
potential increments in the complex potential plane, thusavoiding the need to perform particle tracking using time-stepping algorithms.CVBEM can also be used with complex flux boundary
conditions to continuously define the complex flux withinthe flow domain. This formulation couples the conserva-tion equation, r �~q ¼ 0, with the irrotational assumption,r �~q ¼ 0, which results from specifying both the normaland tangential fluxes. Most specified head and fluxboundary conditions can be readily transformed intocomplex velocity boundary conditions. The complex fluxboundary conditions allows specification both of normalfluxes, as well as flux that is tangent to the boundary. Thus,boundary fluxes oriented at any angle to the boundary canbe considered.By using the complex discharge or velocity as the basic
unknown, CVBEM provides a more accurate flow fieldthan most domain methods because the fluxes or velocitiesare continuous and differentiable functions throughout theflow domain. This results in strict mass conservation andthe ability to avoid the unintentional loss/gain of mass dueto discretization and numerical approximation.The complex approach results in two possible boundary
conditions for every node—either f or c for complexpotential boundary conditions, or qn or qs for complex fluxboundary conditions. Because more than one boundaryvalue can be applied to each node, the problem becomesoverdetermined. Ordinary least squares (CVBEM-OLS)allows for such conditions, improving the accuracy of thesolution. CVBEM-OLS yields a minimum global approx-imation error, providing a more accurate answer. This is incontrast to other methods, such as the TVR scheme, whichmust employ auxiliary boundary conditions to make theproblem well-posed.Three examples related to Toth’s regional flow geometry
are presented. The first uses the original Toth solution,along with two additional solutions proposed by Domeni-co and Palciauskas, and by Nawalany. In both examples,the complex flux formulation provides a solution that is
ARTICLE IN PRESS
Potential Contours Numeric Potential Error
5e005
0.0001 0.0001
0.000150.0002
0.00025
Flux Potential Boundary Conditions Flux Error
0.0001
0.00
01
0.0001
0.000 1
0 .00 02
0.0002
0.000 2
0 .00020.0003
0.00030.000 3
0.00030. 0 00 5
0 .0005 0.0005
0.0005
Flux Flux Boundary Conditions Flux Error
0.0001
0.00
01
0.0002 0.0002
0.0003
0.0005
(a)
(b)
(c)
(d)
(e)
(f )
Fig. 9. Nawalany numeric solutions for (a) complex potential, y ¼ fþ {c, (b) complex flux, q ¼ qx þ {qy, calculated using boundary potentials, yb,
and (c) complex flux calculated using boundary fluxes, qb. Contours of error norms between the analytic and numeric solutions are shown in (d)–(f).
Table 1
Comparison of four solution methods using Nawanaly boundary conditions: Analytic: analytic solution [26], CVBEM : this study, TVR: transport velocity
representation [27], and FEM : finite-element method [27]
y Method x ¼ 0:25 x ¼ 0:50
qx qy qx qy
0.3 Analytic �0:04587 �0:04757 �0:06881 0.00000
CVBEM �0:04584 �0:04745 �0:06865 0.00000
TVR �0:04529 �0:04862 �0:06763 0.00000
FEM �0:04360 �0:05050 �0:06320 0.00000
0.5 Analytic �0:02577 �0:02405 �0:03705 0.00000
CVBEM �0:02573 �0:02401 �0:03698 0.00000
TVR �0:02539 �0:02447 �0:03646 0.00000
FEM �0:02440 �0:04757 �0:03440 0.00000
0.7 Analytic �0:01527 �0:01130 �0:02169 0.00000
CVBEM �0:01524 �0:01129 �0:02165 0.00000
TVR �0:01503 �0:01148 �0:02135 0.00000
FEM �0:01444 �0:01200 �0:02000 0.00000
T.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–1044 1043
accurate, while at the same time being easier to discretize.The use of flux boundary conditions improves the accuracyof velocity distributions, especially near the boundary,which is an important asset in two-dimensional contami-nant transport problems.
This method can also be generalized to include internalsources and sinks, which are first removed from theproblem domain, and then by adding cutting lines betweenthe external boundary and the internal boundary to makethe domains simply connected. Because the velocity
ARTICLE IN PRESST.C. Rasmussen, G. Yu / Engineering Analysis with Boundary Elements 30 (2006) 1030–10441044
components must be single-valued—unlike stream func-tions, which can be multi-valued—the integral along thecutting lines cancel, providing a unique solution to theproblem.
Heterogeneous media can also be considered by deli-neating zones of homogeneous materials. Each zone islinked to neighboring zones along their interfacial bound-ary using flux boundary conditions, which are oriented inopposite directions for each boundary. A further potentialapplication includes transient flow cases, such as forincompressible fluids in non-deformable media, andperiodic boundary conditions [28,29]. Additional applica-tions to Dupuit flow in regional groundwater flowproblems are possible, as well as two-dimensional flow inopen channels.
References
[1] Pollock DW. Semianalytical computation of path lines for finite-
difference models. Ground Water 1988;26(6):743–50.
[2] Cordes C, Kinzelbach W. Continuous groundwater velocity fields
and path lines in linear, bilinear, and trilinear finite elements. Water
Resour Res 1992;28(11):2903–11.
[3] Segol G, Pinder GF, Gray WG. A Galerkin-finite element technique
for calculating the transient position of the saltwater front. Water
Resour Res 1975;11(2):343–7.
[4] Russell TF, Wheeler MF. Finite element and finite difference
methods for continuous flows in porous media. In Ewing RE, editor.
The mathematics of reservoir simulation, vol. 2. Philadelphia PA;
SIAM: 1983, p. 35–106.
[5] Zijl W. Finite element methods based on a transport velocity
representations for groundwater motion. Water Resour Res
1984;20(1):137–45.
[6] Zijl W, Nawalany M. Natural groundwater flow. Boca Raton, FL:
Lewis Publishers; 1993.
[7] Frind EO, Matanga GB. The dual formulation of flow for
contaminant transport modeling: 1. Review of theory and accuracy
aspects. Water Resour Res 1985;21:159–69.
[8] Liggett JA, Liu PLF. The boundary integral equation method for
porous media flow. London: George Allen & Unwin; 1983.
[9] Brebbia CA, Telles JCF, Wrobel LC. Boundary element techniques.
New York: Springer; 1984.
[10] Hunt B, Isaacs LT. Integral equation formulation for ground-water
flow. American Society of Civil Engineers 1981;107:HY10.
[11] Hromadka II TV, Lai C. The complex variable boundary element
method in engineering analysis. New York: Springer; 1987.
[12] Yu GQ, Rasmussen TC. Application of the ordinary least
squares approach for solution of complex variable boundary
element problems. Int J Numer Methods Eng 1997;40(7):
1281–93.
[13] Rasmussen TC, Yu GQ. A complex variable boundary-element
strategy for determining groundwater flownets and travel times. Adv
Water Res 2003;26(4):395–406.
[14] Bear J. Dynamics of fluids in porous media. New York: Dover
Publications; 1972.
[15] Brebbia CA. The boundary element method for engineers. New
York: Wiley; 1978.
[16] Strack ODL. Groundwater mechanics. New York: Prentice-Hall;
1989.
[17] Haitjema HM. Analytic element modeling of groundwater flow. San
Diego, CA: Academic Press; 1995.
[18] Strack ODL. Theory and applications of the analytic element
method. Rev Geophys 2003;41(2):1005.
[19] Baily RT, Hsieh CK. A quadratic-element formulation of the
complex variable boundary element method. Int J Numer Methods
Fluids 1992;14:841–3.
[20] Kassab AJ, Hsieh CK. Application of the complex variable boundary
element method to solving potential problems in doubly connected
domains. Int J Numer Methods Eng 1990;29:161–79.
[21] Lanczos C. Linear differential operators. New York: Van Nostrand;
1961.
[22] Jankovic I, Barnes R. High-order line elements in modeling
two-dimensional groundwater flow. J Hydrol 1999;226(3–4):
211–23.
[23] Rasmussen TC, Yu GQ. Determination of groundwater flow
velocities using complex flux boundary conditions. In: Miller
CT, Farthing MW, Gray WG, Pinder GF, editors. Proceedings of
the XVth international conference on computational methods in
water resources. Chapel Hill NC. New York: Elsevier; 2004.
p. 393–404.
[24] Churchill RV, Brown JW. Complex variables and applications. New
York: McGraw-Hill; 1984.
[25] Toth JA. A theoretical analysis of groundwater flows in small
drainage basins. J Geophys Res 1963;68(16):4795–812.
[26] Domenico PA, Palciauskas VV. Theoretical analysis of forced
convective heat transfer in regional groundwater flows. Geologic
Soc Am Bull 1973;84:3803–14.
[27] Nawalany M. Numerical model for the transport velocity representa-
tion of groundwater flows. In: Sa da Costa A, Baptista AM, Gray
WG, Brebbia CA, Pinder GF, editors. Proceedings of the VIth
international conference on finite elements in water resources. Lisboa,
Portugal. New York: Springer; 1986. p. 251–60.
[28] Bakker M. Transient analytic elements for periodic Dupuit–
Forschheimer flow. Adv Water Res 2004;27:3–12.
[29] Rasmussen TC, Haborak KG, Young MH. Estimating aquifer
hydraulic properties using sinusoidal pumping at the Savannah River
Site, South Carolina, USA. Hydrogeol J 2003;11:466–82.
[30] Lesnic D, Elliott L, Ingham DB. Boundary element methods for
determining the fluid velocity in potential flow. Eng Anal Bound
Elements 1993;11:203–13.