RESISTIVITY MODELLING FOR ARBITRARILY

SHAPED TWO-DIMENSIONAL STRUCTURES *

BY

A. DEY and H. F. MORRISON **

ABSTRACT
DEY, A., and MORRISON, H. F., 1979, Resistivity Modelling for Arbitrarily Shaped
Two-Dimensional Structures, Geophysical Prospecting 27, 106-136.

A numerical technique is developed to solve the three-dimensional potential distribution

about a point source of current located in or on the surface of a half-space containing
arbitrary two-dimensional conductivity distribution. Finite difference equations are
obtained for Poisson’s equations by using point- as well as area-discretization of the
subsurface. Potential distributions at all points in the set defining the half-space are
simultaneously obtained for multiple point sources of current injection. The solution is
obtained with direct explicit matrix inversion techniques. An empirical mixed boundary
condition is used at the “infinitely distant” edges of the lower half-space. Accurate
solutions using area-discretization method are obtained with significantly less attendant
computational costs than with the relaxation, finite-element, or network solution tech-
niques for models of comparable dimensions.

INTRODUCTION

The interpretation of electrical resistivity data has, until recently, been

commonly done with the assumption of horizontal layer stratification. The
recent developments of data acquisition techniques with high accuracy and
substantially higher rates over larger areas warrant a more sophisticated
interpretation of the geologic structure. During the last several decades,
analytical and analog models have been developed for simulating contacts,
dykes, and other simple two and three-dimensional inhomogeneities (Apparao,
Roy, and Mallick rg6g ; Van Nostrand and Cook 1966, McPhar Geophysics
1967, Dey 1967). In most analog modelling, however, only asymptotic behavior
for very large conductivity contrasts is considered for a restricted suite of
physical dimensions. The resistivity response of inhomogeneities of more
general shapes and conductivity have been obtained numerically by Jepsen
(rg6g), Aiken, Hastings, and Sturgul (rg73), Coggon (1971) and Madden (1967,
1971). Finite difference evaluation of the response of two-dimensional structures

* Received January 1977.

** Engineering Geoscience and Lawrence Berkeley Laboratory, University of Califor-
nia, Berkeley, California 94720.
 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES 107

due to a point current source was made by Jepsen (1969) and due to a uniform
field by Aiken et al. (1973). These techniques solved the problem by using the
classical successive-point over-relaxation method due to Southwell (1946).
Madden (1967) used a network solution technique to evaluate the potential
distribution near a two-dimensional conductivity excited by a point current
source. The same problem was solved by Coggon (1971) using a finite element
formulation.
In this paper, a direct and explicit finite difference technique is employed
to solve for the potential distribution in or on the surface of a half-space with
arbitrary two-dimensional distribution of conductivity, due to a point source
of excitation. Two dimensional structures defined here are geologic bodies of
arbitrary cross-secion with infinite extent along strike. The finite difference
scheme is chosen because of the inherent simplicity of the approximation forms
which are also easily amenable to Dirichlet, Neumann, or mixed boundary
conditions. In contrast to the relaxation techniques, a matrix method is
employed to obtain a direct stable solution of the difference equations obtained
by approximating Poisson’s equation over an irregularly spaced, rectangular
grid. The matrix technique has the inherent advantage that the potential
distribution in the entire half space under consideration can be found simul-
taneously for multiple source injection points with attendant computational
costs that can be 10-50 times less than some of the relaxation techniques
hitherto used. As is indicated later, the algorithm developed in this work is
also considerably more efficient than finite element and network solution
techniques for comparable scales of model simulation in terms of central
processor time and core-space required.

FUNDAMENTAL RELATIONS
Ohm’s law relates the current density J to electric field intensity E and an
isotropic conductivity G by
J = GE.
Since stationary electric fields are conservative,
E=-V#.
Hence,
J = - oVc$.
Applying the principle of conservation of charge over a volume, we obtain,
using the equation of continuity,

V . J = ; 6(x) 6(y) 6(z),

108 A. DEY AND H. F. MORRISON

where p is the charge density specified at a point in the Cartesian x-y-z space
by the Dirac delta function.

Equation (I) can be rewritten for a generalized three-dimensional space as

-v * [f4%Y,4 W%Y,41 I= 2 Qs) Ws) Q,), (2)

where (x,, ys, 2,) indicate the coordinates of the point source of charge injected
in the x-y-z space.

By application of elementary vector calculus, equation (2) can be written as

Vdx, y, 2) . Vqq% y, 2) + b(X, y, 4 02$b(X>y, 2) = - 3 6(x,) S(ys) q%), (3)

where

is the Laplacian operator in three dimensions.

If we make the assumption that there be no change in the conductivity

distribution in the y (strike) direction, i.e.,

equations (2) and (3) can be rewritten as

-v . [0(x,2)04(x,y, z)] == 2 Qs) S(ys) Vs), (4a)

and

Vo(x, 2) . vf#J(x, y, x) + 0 (x, x) V2# (x, y, 2) = - 3 W8) %ys) Vd. (4b)

Using the vector relation

vo . vd, = ; [- crwp + V‘+q5) - f#V%]

and substituting this relation in (4b) we get

V2{o(x, 2) 4 (x, y, 4) + 0 (% 4 v24 (% y, 4 - 4(x, Y, 4 V24% 4

(4c)
 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES 109

In equations (4a) and (4c), the potential4 and the source term 2 6 (x8) 8 (ys)
6(z,) are functions of x, y, and z and the conductivity cr is a function of x and z.
For computational ease, it is preferable to solve these equations in Fourier
transformed space (x, K,, z) by transforming y into the K, domain. This
transformation is performed in the forward and backward direction by the
equations

fix, K,, 4 = i f(x> Y, 4 ~0s (K, Y) dy (54

and

.f(x, y, z) = ; i jx, K,, z) ~0s (K, y) d&,, (5b)

0
where f(x, y, z) and f(x, K,, z) are assumed to be even functions of y.
By the transformation (5a) the three-dimensional potential distribution
4(x, y, z) due to a point source at (x,, ys, zs) over a two-dimensional conduc-
tivity distribution cr(x, z) is reduced to the two-dimensional transformed
potential &XC,K,, z) which is a solution of the transformed equation (4a)
-V . [+, 4 V&x, K,, 41 + Ki+, z)&, KY, 4 = @%) a(~). (64
Similarly we obtain from (4~)
V2{+, z) &x, K,, z,} + 4x, 4 V2&, K,, 4 - &, Kg, 4 V2+, 2)
- ~K;o(x, z) q&x,K,, z) = -z&) 6(z,) w4
for a fixed value of K,. The parameter Q” defined in the above equations is
the constant steady state current density in (x, K,, z) space, given by

Qqx,) 8(zs)= i ; 6(x,) 6(z,).

The current density 4 can be related to the current I injected at (x,, z,) by

a=’
zAA ’
where AA is a representative area in x-z plane about the injection point (x,, z,).
The object of this paper is to obtain numerical solutions to equations (6a)
and (6b) subject to proper boundary conditions. These boundary conditions
are :
(I) 4(x, y, z) must be continuous across each boundary of the physical
property distribution of a(x, z), and
110 A. DEY AND H. F. MORRISON

(2) the normal component

each boundary.
of J
(
G B-
a? ) must also be continuous across

I>
The solution of 4(x, K,, z) is obtained by deriving the “difference equations”
of (6a) or (6b), by proper discretization of the (x, K,, z) space over which the
problem is to be solved. Equation (6a) is amenable to a volume discretization
(equivalent to area discretization in x-z space) and equation (6b) is well suited
for a point discretization formulation.

Discretixation of the Two-Dimensional Resistivity Problem

The generalized form of equations (6a) and (6b) can be written as

+ 0(x,4 i(x, K,, 4 = f(x, 2); (x, 4 ~8 (7)

defined in a set % which is assumed to be closed and connected, to have a
non-void interior and to have a sufficiently regular boundary I? with outward
normal q on which

34
a(% 2) i + P(% 2) r = f2(% 2) ; (x, 2) E I‘. (8)

We also have the functions P and f that are at least piecewise continuous
in % and its closure and which satisfy
P(x,z) >oandcr(x,z) >o; (x,z)E%
and
4% 2) 2 0, P(% 2) 2 0;
cc+@ >o; (x,z)Er.

Equation (7) is a self-adjoint, strongly connected and nonseparable elliptic

equation of second order (Varga 1962). The procedure considered in this paper
solves, numerically on a non-uniform rectangular mesh, the problem
L? = -V . [0(x, 2) 041 = f(x, 2) on 3
subject to the boundary conditions (8). The positivity of B(X, x) implies that
the operator L is positive definite.
To define the semi-infinite lower half-space with arbitrary conductivity
distribution, the set 3 is designed with artificial boundaries simulating infilz-
itely distant planes in the horizontal (x-direction) and vertical (z-direction)
extent. Such a lower half-space is indicated by the grid shown in figure I.
 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES III

i= I 2 3 4 N-l N

AX
j= I AXI Ax2 A*3 N- I
2
3
4
5

M-l

Fig. I, Rectangular discretization grid

I
2

The grid is chosen to be rectangular with arbitrary, irregular spacing of the

nodes in x and z direction, respectively. The nodes in the x-direction are indexed
by i = I, 2,3, . . . , N, and the nodes in the z-direction are indexed by j =
I, 2, 3, * * *, M. The left and the right infinite edge of the lower half-section
are simulated by the lines i = I and i = N, respectively. The bottom edge
at infinity is represented by the line j = M. The primary potential due to a
point source as well as the secondary perturbational potentials due to con-
ductivity inhomogeneities in the lower half space fall inversely with distance
in x, y, z space and as Ko(K, . I) in transform space (where KO is the modified
Bessel function and Y the radial distance). Hence, by a choice of large numbers
for M and N with suitably coarsening of the grid as i + I and i --f N and
j + M, and the application of appropriate boundary conditions, the infinite
edges could be simulated by a finite choice of M and N. The representative
equations (6a) and (6b) are applied at any node (i, j) to represent an approxima-
tion over an area AA which is illustrated by the hatched portion in the grid.
It can be seen that for a source point in the interior

AA = (Aa + AZ, - I) . (Aq + AZ,-I)

4
II2 A. DEY AND H. F. MORRlSON

and in the limit, at the ground surface with z ---f o,

@xc + Axt - 1) . AZ,

AA z
4

BOUNDARY CONDITIONS APPLIED ON THE EDGE I' OFTHE REGION%

Since the simulation of the medium is restricted to the conductive lower

half-space alone in 8, it is required that the boundary conditions be specified
at points (x, z) EI’u%. At the ground surface with z = o, this is implemented
by applying the Neumann type condition

The termination of the lower half-plane at x = f GOand z = co is done

by extending the meshes far enough away from the sources and conductivity
inhomogeneities such that the total potential distribution at these edges
approaches an asymptotic behavior. The boundary values along these “infin-
itely” distant edges can be specified from known solutions of the background
homogeneous or layered primary distribution of conductivity. Inhomoge-
neities are viewed as perturbations over this distribution. The boundary
conditions along the left, right, and bottom edges thus become Dirichlet-type.
In the general case of simulation of more complicated conductivity distribu-
tions, often a suitable primary model solution cannot be analytically computed
in the (x, K,, z) space. In such cases, either (i) the potentials at these edges
are assumed to be zero (Dirichlet condition) or (ii) at the edges J&-/I is assumed
to be zero (Neumann condition). It is often found that the first assumption
causes an undershoot and the second assumption causes an overshoot in the
numerically evaluated potentials at some distance from the point source
(Coggon 1971) against the analytical solutions.
We propose to use a mixed boundary condition along the distant left, right,
and bottom edges using the asumptotic behavior of 4 and &,+Y~ expected at
large distances from the source. If the observation point along these edges is
far enough away from the source points as well as from .the conductivity
inhomogeneities, the potential $ in (x, K,, z) space has the form

&, K,, z) = AKo(K, . Y)

where K, is the spatial wave number, I is the radial distance from the source,
A a constant and KO is the modified Bessel function of order zero.
 RESISTIVITY MODELLING FORTWO-DIMENSIONAL STRUCTURES 113

Hence,

= - AKyK1(Ky . I) . cos 8,
where 0 is the angle between the radius vector e, and the outward normal q.

We can therefore write

3&x> K,, 2)
+ ai&, K,, 4 = 0
3T
with
K,KI(K, . I)
a = Ko(Ky . r) ’

Such a mixed boundary condition takes advantage of the physical behavior

of the potential at the distant edges and does not require an a priori assumption
of the nature of the functions 4 and &j/33 that are to be evaluated in terms
of primary models. It also has the inherent advantage of reducing the amount
of coarsening of the grid required as these edges are approached and the reflect-
ions due to virtual sources along the edge nodes are simultaneously eliminated.
In applying this boundary condition the radial distances to all of the edge
nodes may be evaluated from the central point on the top-surface of the mesh.
While for different source locations the corresponding radial distances are
slightly different, in the asymptotic limit used to define a no substantial error
arises from this assumption. It is found experimentally that this mixed bound-
ary condition produces a solution for C$at the edges of the grid that allows a
considerably better fit to the analytically computed solution.

DERIVATION OFTHE FINITE DIFFERENCE EQUATIONS

A. Discretization by Points
At any node in the set %, 4(x, K,, z) must satisfy (6b)

V2(~(~, z&x, K,, 2)) + +, z) V$(x> &,, 2)

- &x, K,, z) V20(x, z) - zK;o(x, z) q&v, KY, z)
= - q&x,) 6 (zs).
It is seen that the two-dimensional partial differential operator

v2 z ($ + 2)

is applied on I+, z), F(x, K,, z) and {cr(x, z) . 4(x, K,, z)}.
Geophysical Prospecting 27
114 A. DEY AND H. F. MORRISON

The known physical property distribution O(X, z) is discretized at each node

by O~J and in the numerical solution a discrete set $i,j at each node is to be
evaluated. At any node (i, j) with irregular grid-spacing in x- and z-direction
the V2 operator on any distribution of PZJ, could be approximated by the
finite difference equation as

v2pi~j = (Axe +‘AQ *

(Pi-l,j -

AX$-1
Ptj) + (Pt+1,1 --P&j)

Axi 1
The application of the difference operator in equation (9) to equation (6b)
for any interior node (i, j) of the grid results in the discretized form given
below.
- z(m-1,j + %,i)
&‘J 1
[ (Axz + Axr-l) . Axz-1
- Z(Q,j-1
+Oi,i)
+ hj-’ 1
(Aq + AZ,-1) . A+1

(Axz + Axr- 1) . Axi _ 1 + (Axz + Axi _ 1) . Axi

+
qqj-1 + O&i)
(Azj + Az~-~) . Az~-~ +
2(qj+1

(AZ, + A+r)
+ %,j)

. AZ, + 2Kiai~j 1
= 2Q8(Xs)
kqz,). (10)
In notational form, equation (IO) can be written as

-
2(%-l,i + O&j)
(11.1)
where c2 = (Axz + Axz -1) . Axz _ 1’
is the coupling coefficient between the nodes (i, j) and (i - I, j),

- 2(%+1,j + Q,j)
(11.2)
cg= (Axz+Axz-1) . Axz
is the coupling coefficient between the nodes (i, j) and (i + I, j),

- z(m,j-1 + O&j)

‘$ = (AZ, + As-r) . A+,’ (11.3)

is the coupling coefficient between the nodes (i, j) and (i, j - I),
 RESISTIVITY MODELLING PORTWO-DIMENSIONAL STRUCTURES 115

2(0i,3+1 + Q,j)
@ = -I
B (AZ, + A.+i) . AZ,’ (11.4)

is the coupling coefficient between the nodes (i, i) and (i, j + I),

and

c$ = - [Cf+ cg + cy + cg- 2K;ai,j],

is the self coupling at node (i, i).

The difference equation (II) thus obtained indicates that the solution of 3
at node (i, j) is dependentonly on the values of $ at the adjacent nodes (i - I,
i), (i + I,j), and (i, i + I). The coupling coefficients C are func-
6, j - I),
tions of the geometry of the discretization grid and the physical property
values O~J, and hence known at all nodes in the set 3.
It should be noted that in the finite difference approximation to the operator
V2 it is implicitly assumed that the functional distribution of $, o $/3x and
0 &#/&z are at least piecewise continuous. Equation (II) is valid for any distribu-
tion of O~,J(O< 0 < CO)and so, for any interior point in the grid the required
boundary conditions on the continuity of 4 and 0 $/JYJ are satisfied across
any of the rectangular element boundaries.
While the coefficients and the self-adjoint form of equation (II) are valid
for any interior node, the corresponding equations are somewhat altered for
the nodes located at the top surface, and left, right, and bottom edges of the
grid. The difference equations (ra)-(Sa) for these nodes, with appropriate
mixed boundary conditions based on the asymptotic behavior of potentials,
are formulated in the appendix A.

B. Discretization by Area

At any node in the set 3, the constitutive relation for the unknown potential
&x, K,, z) is given by the self-adjoint eliptic partial differential equation (6a)
as

-V . I+, 4 V&G K,, 41 + +J(K 2)kc K,, 2)

= @(xJ 6(z,) with (x, z) ES

and the boundary condition given by equation [8), such that

with u > o, p 2 o, and cc+ p > o, prescribed on the boundary I’u%

116 A. DEY AND H. F. MORRISON

The physical property distribution ai,5 at any node (i, j) of the rectangular
grid, described in the previous section, could be discretized in the sense that
or,g now indicates the conductivity in a region bounded by the nodes (i, j) and
(i + I, j) in the x-dr‘rection and the nodes (i, j + I) and (i + I, j + I) in the
z-direction. The numerical solution of (6a) that consists of a discretized set
of &J at each node, is to be evaluated. As in the previous section, the node
(i, j) is assumed to represent the closed mesh region AAr,j about the node, as
illustrated in figure I. It is seen that for a nodal point in the interior one has
AA _ (Axi + h-1) (AZ, + b-1)
w -
4
and the limit z + o, - a nodal point on the ground surface -
AA _ (A& + Axz- 1) . Azj
u -
4
For each node (i, j) for which &J is unknown, we now integrate equation
(6a) over the corresponding mesh region AAr,j to obtain

I
Using the relation 0 = -
zAAa,j

(13)

Using Green’s theorem

ss
we obtain
a$-
A*c,i V . (c$) da =
%f
G -3~ dl, (14)
 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES 117

where q is the outward normal direction and L~,J is the contour line enclosing
the mesh region AA r,j. It is seen from equation (14) that over every element
of % and on the boundary I’ the boundary conditions given by equation (12)
could be directly implemented in the first term of the discretized equation (13).
The detailed description of the distribution of oa,j, the representative mesh
area AAz,j, and the line contour Li,* about a node (i, j) in the interior of the
grid is illustrated in figure 2. The first term on the left hand node of equation
(12) is given by

(15)

i-. I,j-I
. i+l , j-l
..*
VIII i
/
,d / ii
. .
i-l, j i,j b
I i+l, j
c AAi,j
. L
V iv

. . .
i-l,j+ I i,j+l itl, jtl
Fig. 2. Discretized area element AAt,,.

The contour integration along the line LQ is subdivided into eight sub-
sections as indicated in figure 2. Integrating along the entire path Lgj we get
by approximating i$/~r by central difference

A.zj . or,j
--.-
+ 2
118 A. DEY AND H. F. MORRISON

A.q-I . oi-l,j-1

b-hi-4v Axt-I . Oi-l,j-1

AXz-1 1 -+
2

(16)

Similarly, the second term on the left hand side of equation (12) can be
expanded as

ai-l,f-l . AQ-I . A.+1

4
o~,~ . Axi . AZ, o~-I,~ . Axz-I--- . A5
+ + = A(ot,j, &,j) * &,j. (17)
4 4 1

Substituting the difference approximation (16) and (17) in equation (13),

we obtain for an interior node point (;, j)

where Cz = -
AZ,-,
. o~-I,~-I + Aqoi-l,j
2Aq-1 1
is the coupling coefficient between nodes (i, j) and (i - I, j),
(18.1)

c"a' = -
[
AZ,-, . qj-I

is the coupling coefficient

+ AZ, . oi,g
2Axz 1
between the nodes (i, j) and (i + I, j),
(18.2)

Axg-I . o~-I,~-I + Axz . q-1

c$ = - (18.3)
2Azj-I] I
is the coupling coefficient between the nodes (i, j) and (i, j - I),
 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES II9

=i-1,~ + Axr . oz,j

is the coupling coefficient

2A.q 1
between the nodes (i, j) and (i, i + I),
(18.4)

and
(18.5)
is the’ self coupling coefficient at node (i, J).

The self-adjoint difference equation (18) indicates that the solution of 4 at

(i, i) node is dependent only on the values of 6 at the adjacent nodes (i - I, j),
(i + ~,j), (i,j-I) and (i,j + I). Th e coupling coefficients are known func-
tions of the geometry and the physical property distribution in the set 3.
The application of the boundary conditions on l? is made such that on the
ground surface (z = o), the equation (12) has cc= o, p = ot,g, and f~ = 0.0.
On the left, right, and lower edge of I’, simulating the “infinitely” distant
boundaries, equation (12) is applied with ct = KyKl(KZ/ . r)/Ko(K, . Y), p = I,
f~ = 0.0. The modified difference equations (rb)-(8b) obtained upon application
of these conditions at the special nodal points, are developed in appendix B.
Equations (II) and (ra)-(8a) obtained with the discretization by points or
the set of equations (18) and (rb)-(8b) obtained through discretization by area,
can be solved for all &,j by using Successive Point Overrelaxation (Southwell
1946)) Successive Line Overrelaxation (Varga 1962)) Alternating Direction
Iterative Methods (Peaceman and Rachford 1955, Douglas and Rachford 1956,
and Gunn 1964). In these methods, an initial assumed distribution of $2,~ over
the grid is relaxed by successive refinement through iterations. The refinements
in the individual methods are either in terms of individual nodes, successive
rows or columns of nodes or of &,f alternately along a column and a row.
The refinement obtained upon an iteration is further updated by the use of
an optimal overrelaxation factor or by successive use of the Chebychev over-
relaxation acceleration parameter (Concus and Golub 1973). In the large grids
under consideration (1000-2000 nodal points), these techniques require a
minimum of 30-100 iteration sweeps through the entire grid for each location
of the point source of current injection. In addition, the convergence rates of
these iterative techniques are highly dependent on the dimensions of the
grid spacings and the nature of the physical property distribution. Detailed
analysis of these techniques and their comparison are given by Varga (rg62),
Forsythe and Wasow (rg6o), and Mitchell (1969). It is found experimentally
that for the nonseparable, elliptic, self-adjoint difference equations that occur
in the present formulation, direct matrix inversion techniques can be applied
to solve for potential distributions with multiple point source locations, with
120 A. DEY AND H. F. MORRISON

an attendant cost of computation that could be IO to 50 times less than the

iterative techniques mentioned above.

Matrix Formulation
In the matrix formulation, equations (II) and (ra)-(Ba) or the set of equa-
tions (18) and (rb)-(Bb) applied to all nodes (i, j), i = I, 2, . . . , N and j = I,
2, . ..) M may be solved simultaneously. For an example, a 4 x 4 grid is shown
in figure 3.

I 13

6 IO 14
2
7 II
3 I5

4 - .- I6
8 I2
Fig. 3. Sample node numbering convention for rectangular grid.

The individual nodes are numbered I through 16 starting from the left
topcorner and periodically increasing along each column. The set of simul-
taneous equations for all the nodes in the ‘grid can be written in matrix form
(see table I) or, symbolically, as
c$L s. (19)
The MN x MN matrix C is called the Capacitance Matrix and is a function
of the geometry and the physical property distribution in the grid. It is to be
noted that for multiple source locations, the matrix C remains unaltered.
122 A. DEY AND H. F. MORRISON

Hence, only one inversion of C provides the solution to different sets of vectors
4 for the different. source locations that constitute the vector S, through back
substitution. This is an inherent advantage in the direct matrix solution
method over the semi-explicit iterative methods mentioned previously.
The structure of the matrix C is blocktridiagonal, sparse, and banded. The
choice of sequential numbering of nodes along columns or rows dictates the
bandwidth required. In resistivity surveying with large surface coverage a
grid with N > M is usually required. The numbering of nodes along rows
causes the number of co-diagonals to be 2N; if the nodes were to be numbered
along columns, as shown in figure 3, the number of co-diagonals would be ZM
resulting in considerable saving of the core space required in the solution
technique.
The coefficients in the matrix C as indicated in equation (19) are derived
by using the mixed boundary condition along the “infinitely distant” edges of
the mesh in x- and z-directions. If, however, for nodes lying on the left, right,
and bottom edges (i.e., 1-4, 13-16, and 8 and 12, respectively) known potential
values (e.g. primary model potentials) are to be assigned (Dirichlet condition),
this could be simply achieved by making the corresponding row C~J = 6~
and by adding to the corresponding element in the vector S the known Dirichlet
potential for the node.
The capacitance matrix C has the following properties.
(i) C&i > o,i = 1,2, . . . . MN;
MN
(4 ci,i 2 c I C&j 1, i = I, 2, . . .) MN,
i=1
;*i
i.e. C is diagonally dominant;
(iii) C is irreducible and has a strongly connected directed graph (Varga
1962); and

(iv) C possess Young’s Property A (Young 1954).

It is shown by Varga (1962) that the explicit difference equations that give
rise to the matrix C with properties described above are inherently stable for
arbitrary grid spacings.
The matrix C when obtained through discretization by point representation
is found to be nonsymmetric, for unequal grid spacings in + x- and Jc z-direc-
tion and for arbitrary distribution of a(%, z). Discretization by area over an
inequally spaced rectangular grid with arbitrary distribution of B(X, Z) as
indicated by equations (12) and (rb)-(8b), h owever, makes the capacitance
matrix C a positive definite, symmetric matrix.
 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES 123

The solution of the equation C 4 = S, when C is nonsymmetric and banded,

is best done by the well known Gaussian elimination algorithm. For the area
discretization method, when C is symmetric and positive definite, the solution
is best obtained by symmetric Cholesky decomposition (Martin and Wilkinson
1965). This method is based on the following theorem :

If A is a positive definite matrix of order n and banded form such that

acg F= 0 (I i---j I > YM),

then there exists a real, nonsingular lower triangular matrix L such that
LLT = A where LQ = o(i --j > m). (20)

The elements of L may be determined row by row by equating elements on

both sides of equation (20). The solution of the set of equations Ax = b can
be determined in the steps
Ly=b,LTx=y. (21)

This decomposition technique is especially effecti-ve when m < n. There

are then approximately a(~ + I) (m + 2)/2 multiplication and n square roots
in the decomposition, approximately 274~~2 + I) multiplications are involved
in the solution steps, and any number of right hand sides can be processed
when L is known. For symmetric, banded matrices, this operation count is
lower than that for Gaussian elimination methods.

Inverse Transformation of 4(x, K,, z)

The solution of 4(x, K,, z) is outlined in the previous sections. In order to
obtain the solution of # in the (x, y, z) domain, the set of $(x, K,, x) for several
optimal values of K,, chosen to discretize the interval o < K, < co, are
obtained. The inverse Fourier transform is then performed by numerically
integrating equation (5b). The general behavior of #(x, K,, .z) indicates an
asymptotically flat response as K, -+ o and a monotonic fall off to zero as
K, + CO. The integration is performed by fitting the envelope of ,&KY) in
each subsection KY1 _< K, 2 Ku2 by an exponential and using the analytic
form
%z K,Y,
%
s e-aKg cos (K,b) dK, = s . [b sin (bK,) - a cos (bK,)]
KY* *

and taking the cumulative sum of these subsectional integrals up to reasonably

large values of KY to obtain an accuracy of I percent.
124 A. DEY AND H. F. MORRISON

Determination of the Apparent Resistivity Response of 2-Dimensional Structures

In electrical resistivity surveys a current source + I and a current sink - I
are used to energize the conductive earth. A potential difference AV is measured
between two points located at arbitrary azimuthal orientation (for surface
arrays) or colatitudinal configuration (as in down-hole-surface arrays). A
parameter apparent resistivity is defined as a function

pa-G? (22)

where, for the configuration illustrated in figure 4,

G=25i

Cl
+
Fig. 4. Electrode configuration for an arbitrary geometric factor.

For a homogeneous half-space, pa is the true intrinsic resistivity of the

medium. If, however, the lower semi-infinite medium has inhomogeneous
conductivity distribution, pa indicates the resistivity of an apparent homo-
geneous half-space that results in an identical AV for the transmitter-receiver
array under consideration. All interpretations of electrical resistivity work are
done using the apparent resistivity concept described above.
It can be seen that the d.c. potential distribution at any point is a super-
position of the solutions of two point sources of current located at the trans-
mitting electrodes of amplitude + I and - I.

Results
To estimate the accuracy of the two-dimensional resistivity technique
described in the previous sections, a two-layered earth model was simulated.
 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES 12.5

The resistivity of the top layer of thickness IOOO m was assumed to be IOO Qm
and that of the bottom layer to be IO Qm. A collinear dipole-dipole array was
deployed in a sounding mode with a dipole length of IOOO m and with dipole
separations N = I, 2, . . ., 15. The results of such a numerical simulation are
indicated in figure 5 with circles, and the analytically computed response for
the two layered model is shown by the curve with solid line. It can be seen
that the numerical results approach the analytic solution with an accuracy
of better than 5%.

ANALYTIC
000 0 NUMERICAL

DIPOLE - SEPPRATION , N -

Fig. 5. Comparison of numerical and analytic results for a dipole dipole sounding over a
two-layer model.

The lrectangular grid used in’the simulation of the above model consisted of
113 nodes in the x-direction and 16 nodes in thez-direction. The central 103 nodes
were equally spaced and finely described. The node spacing approaching the left
and the right edge were rapidly coarsened to simulate the infinitely distant
edges. The vertical resolution of the node distribution was similarly fine near
the ground surface and coarsened considerably to approach the “infinite depth”
boundary. In the result presented, the discretization by area method was em-
ployed with the empirical mixed-boundary condition applied at the edges.
In another example, illustrated in figure 6, a routine interpretation over a
complex structure is done by obtaining a “best fit” to the apparent resistivity
data observed in field survey, using the collinear dipole-dipole array, dis-
played in a standard pseudo-section plot.
 126 A. DEY AND H. F. MORRISON

GRASS VALLEY
LINE A-A’
NORTH LEACH HOT SPRlNGS SOUTH
5.5 50 45 40 35 I 30 25 20 15 IO 05 KM
I I I I 1, 1 I I I I I I, 1 I

OBSERVED DATA
APPARENT RESISTIVITY IN fi m

INTERPRETED DATA
APPARENT RESISTIVITY IN i-l m

NOW” SOUTH
55 50 45 40 35 3.0 25 20 1.5 IO 05
1 I I I I I I I I, I I I I I I I I I I I IKL
30RM I2O.Q M 20Rb.4 w
3oi-h
_ IOn
4&Q-M
05- 05

Zs2M KM

“O 1 I 20Cl-M 10

1.51
INTERPRETED SECTION

Fig. 6. Interpretation of an apparent resistivity pseudo section.

LkC’l,LSSiO~

In most of the standardized use in our interpretational task for B collinear

dipole-dipole array, a grid of 113 x 16 (= 1808) nodes is used. The.solution
at each of these 1808 nodes is obtained for 23 transmitting point somces., The
discretized problem is solved in our algorithm by obtaining repeated iets of
 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES 127

solution obtained for five optimally chosen K, values. The CP-times required
to solve for an arbitrary z-dimensional structure are 41 seconds and 24 seconds
for discretization by point and discretization by area methods, respectively,
on a CDC 7600 machine. An identical problem when solved using a network
solution technique to the same accuracy requires about 35 seconds of CP-time
using comparable core storage. A finite element algorithm available to us
could not be compared for an equivalent extent of the pseudo section evalua-
tion, due to lack of core space to fit the problem. However, from comparison
of problems of smaller dimensions, the new technique is estimated to be
considerably faster and more economic in terms of cost and storage require-
ments than the finite element algorithm (Coggon 1971).

ACKNOWLEDGEMENTS
The authors are indebted to Mike Hoversten for computational assistance.
Support for this work has been provided by the United States Energy Research
and Development Administration through Lawrence Berkeley Laboratory.

APPENDIX A
DIFFERENCE EQUATIONS FORTHE BOUNDARYNODES USING POINT
DISCRETIZATION SCHEME
a. For modes located 0% the line z = o
For all nodes (i, j) with i = 2, 3, . . . , ii1 - I, j = I, the boundary condition
is of the Neumann type, i.e.

This is implemented by assuming a fictitious row of nodes in the air at j = o,

such that the potential #Ji,z and conductivity ci,2 at nodes (i, 2) are reflected
at the imaginary nodes (i, 0). This assumption leads to the difference form of
equation (6b) given by

where the coupling coefficients are given by

(r.ra)

(r.2a)

(I.34
128 A. DEY AND H. F. MORRISON

and
C$ = - [Cz + C$ + C$ - 2K;a,,,]. (I.44

b. For the top, left, and right corner nodes on line z = o

The nodes (I, I) and (N, I) which lie on the “infinite” boundary, the normal
component of J in z-direction is zero (Neumann condition). The x-component
of J, however, satisfies the mixed boundary condition

b&
G + ct$ cos 0 = o with a =
K&l Wd
Ko(Kr’) ,

where 0 is the angle between the radius vector I from the source point and the
outward normal in x-direction. If we assume a set of fictitious column of
nodes in the lower half-space at i = o and i = N + I, then the potential at
these nodes could be expressed entirely in terms of the potential on the boundary
node (i, j) through the mixed boundary condition.
These conditions applied to the equation (6b) result in the difference equa-
tions, for the top left node (I, I) :

where
(y = - (w + az+v)
R > (2.Ia)
(Aa) 2

2=2,j
C$ = - [C$ + Cg - 2K;qr] - cos O), (2.34
(Ax()’ (I
and for the top right node (N, I) :

(3a)
where

(3.14

(3.24
and

c$ = - [Cz + Cg - 2K;cz,,] - $3 (I - aAxaml cos 0). (3.34

 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES 129

c. For the nodes located on the bottom edge of the mesh (z -+ co)
At the nodes (i, M), i = 2, 3, . . . , N - I, that are on the lower “infinite”
boundary, the application of the mixed boundary condition described above
results in the difference equation

cy J L,* + cg &+l,j + q &,1-j + q &2,j = 2Q~(X,) @,I (4a)

with the coupling coefficients

(4.14

c”” = - (w-1 + 4
T (4.34
(AZ, _ I) 2
and

C$ = - [C; + Cj$ + C$ - 2K;qj] - iA$3 . (I - - crAz$-, . cos ‘3.

(4.44

d. For the bottom left and right corner nodes

For the nodes (I, M) and (N, M) the components of Jz and Jz satisfy the
mixed boundary conditions

4
& +4 cos 81 = 0

and

Ji
,+a4 cos e2 = 0,

where 01 and 0~ are the angles between the radial distance Y from the source
to the node (i, j) and the outward normals in the x- and z-directions, respect-
ively.
The application of these boundary conditions results in the difference
equation for the node (I, M) given below:
cg ${ + 1>, + c$ $I >j - 1 + c$! h,, = z@(x,) Q,), (54
where

cg = - h+l,j + O&j) (5.14

(W
Geophysical Prospecting 27 9
130 A. DEY AND H. F. MORRISON

@j = _ (%-1 + %i)
T (AQ-1)2 ’ (5.24
and

C$! = - [Cg + C$ - zK;r& - z2 (I - crAxa cos 0,)

- s2 (I - UAZ~-I cos 0,) (5.34

For the node (N, M) the difference equation is

where
pj = _ (Q-U + %I)
L (6.ra)
(Axi-,)2 ’

(6.za)

C$ = - [Cy + C; - zK&] - -.)

(A;;;)2 (I - aAXi-l cos 0,)

- 202,g (I - CXAZ~-I- cos 0,).

(Az~-I)~

e. For the nodes on the left edge of the mesh

The difference equation for the nodes (I, j) for j = 2, 3, . . ., M - I, has
the form given below
cg &+1,j + cy &J-l + G &,j+l + cc i,,, = zQS(x,)
W,L (7a)
where
cg = - h+l,j + CG,,)
(7.14
(4)” ’
+ Q,j)
--2(Q,l-1
q =
[AZ, + A.z~-I) . A.z~-I’ (7.2a)
-
2(w+1 + w)
cg = (AZ, + Az~-I) . AZ,’ (7.34
and

uAix cos e). (7.44

 RESISTIVITY MODELLING FOR TWO-DIMENSIONALSTRUCTURES r3r

f. For the nodes on the right edge of the mesh

The difference equation for the nodes (N, j) for j = 2, 3, . . . , M - I, that
are on the edge corresponding to x ++ + co, is given below
cy &l,j + c$ &j-l + C% ?t,j+l + C$ ~4,~= zQW,) Q,), (84
where
- (061,j + w)
I (8sa)
(&-I)~

C$j= (8.za)

- qot,i+1+ G,j)
cg = (8.34
(AZ, + AZ*-I) . AZ, ’
and

c$ = - [CZ + c$ + c$ - zK;a& - St> (I - CCAX,_ 1 cos 0). (8.4a)

APPENDIX B
DIFFERENCE EQUATIONSFORTHE BOUNDARY NODES USING AREA
DISCRETIZATION SCHEME
a. For nodes located on the ground surface (z --f o)
The mesh region AA t,j is enclosed by the coutour L~,J defined by the sub-
sections iii, iv, o, vi, a, b as shown in figure 2. For all nodes (i, I), i = 2, 3, . . . ,
N - I. The finite difference equation is given by

c$f 4,-l,, + q &+l,j + G J&j+1 + cg i&j = f Q,) &a Pb)

where
- A.q . ~-1,~
q = (r.xb)
2Axz-1 ’

(r.zb)

c$= - Ax~-I . cq- 1,j -I- Axr ’ =w

2A.q
W-4

0<-1,~ . AXCLI . AZ, qj . Axz . A.z~

4,,p A&j) = K; + (r.4b)
4 4 I ’
and
C$=-[C~+C~+C$-A(o 1,j’ 4,~)1~ (I.+)
132 A. DEY AND H. F. MORRISON

b. For the top left and right coynev nodes on line z = o

For the nodes (I, I) and (N, I), the mesh region AAs, is bounded by the
contour L,,i with the subsections (iii, iv, c, b) and (v, vi, a, c), respectively.
The component of J in the z-direction satisfies the Neumann condition c . 3#/
az = o, and the component of J in the x-direction satisfies the mixed boundary
condition

ai
& + UJ cos e = 0,

where
K&l KP9
u = Ko(K#)

and 0 is the angle between the radial distance I from the source to the node
(i, j) and the outward normal in the x-direction. The finite difference equation
for the top left corner node then becomes

(2b)

where

(z.Ib)

(z.zb)

kib)

and
AZ, . o~,~. u cos 8
c$ = - [Cg + Cg - A(r+ AiJ] + 2
@4b)

For the top right corner node, we get

(3b)

where
- Azj . GC-I,~
q = W-4
2Axi-1 ’

- Axg-I . 0i-1,~
cg = Cub)
2Azj ’
 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES I33

A$,J= K; b-1,j * 4
Axi-, . AZ,
ACc&*, kid4
and
Az3. . 02m1,, . a cos 0
c$ = -[CZ + P B -A(o ZJJ4,dl + ______ . b-44
2

c. For the nodes located on the bottom edge of the mesh (z -+ co)
The nodes (i, M), i = 2, 3, . . . , N - I, the mesh region AAt,, is bounded
by the contour L~J defined by the subsections i, ii, b, a, vii and viii. The
finite difference equation for any of these nodes is

(4b)

where
- A.zj-l . oz.-l,j-1
cz =
sAxi - I
k.Ib)
- A+, . ai,j-I
cg = (wb)
zhxs ’

A\xi-1 . oi_l,j-1 + Axti . ‘i,j-1

q =- (4.3b)
2A.q - 1

~i-l,~-l - A+, . AZ,-, + ~i,~-l . A%, . A+ (44b)

A bJt,pAi,J = K;
4 4 I
and
c$ = - [Cz + Cg + Ctj - A(qj, A,,J]
Ax,-I . aa-~,~-l + Axi . q-1
+ 2
a cos 8. (4@)

d. For the bottom left and right comev nodes

For each of the nodes (I, M) and (N, M) the representative mesh region
AA~J is bounded by the contour line Li* described by the subsections (i, ii,
b, d) and (d, a, vii, viii), respectively. Applying the boundary conditions, the
difference equation for the node (I, 44) is

where
- AZ,-, . c(,~-I
c$ = Wb)
zaxt ’
134 A. DEY AND H. F. MORRISON

(wb)

(5&J
and

+
A%* O&i-1 cccos01+
A+1
__-____
* q-1
. ucosf32 . (544
2 * 2 )

The difference equation for the bottom right corner node (N, M) is

Pb)
where
- Azi-l . o~-I,~-I
c; = (6.rb)
2Axg _ I

- Axp-I . O$-l,j-1
q = , (6.2b)
2Azj - I

q-1,~-1 . Axi-I . Azi-,

A h,p A&j) = K; (6.3b)
4
and
C$! = - [C$f + C$’ - A@,,,, A&]
Ax~-I . og-l,j-1 Az~-I . at-l&l
+ . cc cos 01 + . a cos 02 . (64)
( 2 2

02 and 01 are the angles between the radial distance Y from the source and
the x- and z-directions, respectively.

e. For the nodes on the left edge of the mesh

For each of the nodes (I, j), j = 2, 3, . . . , M - I, the mesh region AAdj is
bounded by the contour Lij defined by the subsections i, ii, iii, iv, c, and d.
Applying the boundary conditions, we get the difference equation

(7b)
where
q _7 -
Az~ . qg + Azj-1 . q-1
2Axc
(7.Ib)
 RESISTIVITY MODELLING FOR TWO-DIMENSIONAL STRUCTURES 135

AXE . oi,j-l
c$ = - (7-2b)
zA.z-I ’

(744

z . A.z-I oa,j . Axz . A.q

A(o,,,,A,,j)= K; 02J-1. + (74)
4 I ’
and
c~=-[c~+c~+c~-& 6,iJ 4,Jl
Aq . 08,~ + Aq-I . OZJ-I
+ . cc cos 0. (7.9)
2

f. For the nodes on the right edge of the mesh

For each of the nodes (N, j), j = 2, 3, . . . , M - I, the mesh region AA$,j
is bounded by the contour LQ described by the subsections d, c, v, vi, viii, and
viii. The difference equation for any of these nodes is
I
Cy &i-l .j + c$ Ji,j-l + “‘JB &j+l + c$ &,j = ; YxJ WJ~ (8b)
where
Az~ . ~6-1,~ + A+, . og-l,p-I
(y = _ (8.Ib)
L
[ 2Axr r 1 I1

Axe-1 . oi-q-1
c$ = - (kb)
2A.z-1 ’

Axz-1 . oi-l,j
cg=- (@A)
2Aq ’

c-~,~-I. Ax,-I . A+1 ~-1,~ . Axi-I . Aq

A h,p A&j) = K; + (84)
[ 4 4 I
and
C+-[C;+C$+Cg-A((o a,j~ Ai,3)1
AZ, . 0~.m1,~+ Azj-l . oi-l,j-1
+ ) . crcos0. (Ob)
2

KEFERENCES

AIKEN, C. L., HASTINGS, D. A., and STURGUL, J. R., 1973, Physical and computer modeling
of induced polarization, Geophysical Prospecting 21, 763.
APPARAO, A., ROY, A., and MALLICK, K., 1969, Resistivity model experiments, Geo-
exploration 7, 45.
136 DEY, MORRISON, TWO-DIMENSIONAL STRUCTURES

COGGON, J. H., 1971. Electromagnetic and electrical modeling by the finite element
method, Geophysics 36, 132.
CONCUS, P., and COLUB, G. H., 1973, Use of fast direct methods for the efficient numerical
solution of non-separable eliptic equations, SIAM J. Numer. Anal. IO, 1103.
DEY, A., 1967, Model studies on electrical profiling over thick conducting veins, M.Sc.
Thesis, Indian Institute of Technology, Kharagpur.
DOUGLAS, J., JR., and RACHFORD, H. H., JR., 1956, On the numerical solution of heat
conduction problems in two or three space variables, Trans. Amer. Math. Sot. 82,
421.
FORSYTHE, G. E., and WASOW, W. R., 1960, Finite difference methods for partial dif-
ferential equations, John Wiley and Sons, Inc., New York.
GUNN, J. E., 1964, The numerical solution of V . aVu = f by a semi-explicit alternating
direction iterative method, Numer. Math. 6, 181.
JEPSEN, A. F., 1969, Numerical modeling in resistivity prospecting, Ph.D. Thesis, Uni-
versity of California, Berkeley.
McPhar Geophysics, 1967, Catalogue of resistivity and IP model data, McPhar Geo-
physics Ltd., Ontario, Canada.
MADDEN, T. R., 1967, Calculations of induced polarization anomalies for arbitrary two-
dimensional resistivity structure, paper presented at Symposium on IP, University
of California, Berkeley.
MADDEN, T. R., 1971, The resolving power of geoelectric measurements for delineating
resistive zones within the crust, Geophys. Monogr. 14, AGU, p. 95.
MARTIN, R. S., and WILKINSON, J. H., 1965, Symmetric decomposition of positive
definite band matrices, Numer. Math. 7, 355.
MITCHELL, A. R., 1969, Computational methods in partial differential equations, John
Wiley and Sons, Inc., New York.
PEACEMANN, D. W., and RACHFORD, H. H., JR., 1955, The numerical solution of parabolic
and elliptic differential equations, Jour. of Sot. Indus. Appl. Math. 3, 28.
SOUTHWELL, R. V., 1946, Relaxation methods in theoretical physics, vol. I, Clarendon
Press, Oxford.
VAN NOSTRAND, R. G., and COOK, K. L., 1966, Interpretation of resistivity data, U.S.G.S.
Prof. paper 499.
VARGA, R. S., 1962, Matrix iterative analysis, Prentice-Hall, Englewood Cliffs, New
Jersey.
YOUNG, D., 1954, Iterative methods for solving partial difference equations of elliptic
type, Trans. Amer. Math. Sot. 76, 92.