GWPHYSICS, VOL. 41. th0. I (l-tBKtiAKY, 1976). P. 62-78.
I2 FIGS
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
FINITE-DIFFERENCE RESISTIVITY MODELING FOR
ARBITRARILY SHAPED TWO-DIMENSIONAL STRUCTURES
IRSHAD R. MUFTI*
Resistivity surveying is commonly done by us-
ing a point-source dipole. Consequently, a finite-
difference evaluation of apparent resistivity curves
implies the use of three-dimensional simulation
models which necessitate prohibitive computer
costs. However, if we assume variation of resis-
tivity only in two dimensions and use a line-source
dipole for setting up the finite-difference model of
a given structure, the potential field can be evalu-
ated easily.
A discrete version of the resistivity problem in
two dimensions, which takes into account nonuni-
form grid spacing, is presented as a system of self-
adjoint difference equations. Since the iterative
solution of such a system does not require grid
spacing to be less than a certain critical value, it
was successfully used for the development of fast-
convergence finite-difference models. By exam-
ining in detail the characteristics of the matrix as-
sociated with the evaluation of the potential field,
it is demonstrated that the proposed modeling
INTRODUCTIO&
The interpretation of electrical resistivity field
data is commonly done by assuming that the sub-
surface consists of a system of uniform and homo-
geneous parallel layers. Very often this assump-
tion amounts to a gross simplification of the
actual geology. A much more realistic model of
the subsurface can be constructed by making pro-
vision for the variation of resistivity in two direc-
tions. Very often, the error introduced by assuming
no variation of resistivity along the third direction
can be minimized by choosing the direction of the
measurement profile at right angles to the direc-
Paper presentedat the 44th Annual International SEC Meeting, November 12, 1974 in Dallas, Tex. Manuscript re-
ceived by the Editor February 18, 1975: revised manuscript received May 7, 1975.
* Amoco Production Co., Tulsa, Okla. 74102.
0 1976 Society of Exploration Geophysicists. All rights reserved.
62
procedure will remain stable for all conceivable
geometries and resistivity distributions. It was
used for the investigation of certain models for
which the corresponding results could also be
computed analytically. A direct superposition of
results obtained in the two cases shows that they
are virtually identical. By making use of the recip-
rocity theorem, a computational short-cut, which
provides the evaluation of vertical sounding
curves for a line-source dipole in a single step, is
put forward.
Special problems related to the optimization of
acceleration parameters as well as the estimation
of the potential function along the subsurface
boundaries of the model are discussed. It is con-
cluded that by surrounding the model by a termi-
nation strip of very large effective width, either
Neumann- or Dirichlet-type boundary conditions
can be used for simulating a semiinfinite medium
without introducing signficant errors in the re-
sults.
tion of strike. In situations where such a choice is
not possible, this error will not adversely affect the
interpretation of field data because most of the
current supplied to the ground is concentrated in
the zone adjacent to the vertical cross-section con-
taining the current electrodes.
If we further assume that the current is supplied
to the ground by means of a pair of infinite line
electrodes, extending at right angles to the direc-
tion of the measurement profile, the resistivity
problem can be treated as purely two-dimensional
and can be solved easily by a finite-difference
SC heme.
 Finite-Difference
We shall base our treatment on this assumption
of a line-source dipole and shall be concerned
with the finite-difference evaluation of the poten-
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
tial field and apparent resistivity curves for two-
dimensional structures of arbitrary shape. Studies
of this problem along these lines were recently re-
ported by Jepsen (1969) and Aiken et al (1973).
Both have attempted solving this problem by us-
ing the straightforward relaxation method due to
Southwell (1946), which is much too slow even on
modern high-speed computers. Moreover, the ti-
nite-difference scheme presented in both cases be-
comes unstable except for certain resistivity ratios
and specific subsurface geometries. In the follow-
ing pages, fast-convergence finite-difference sys-
tems based on the stability analysis of the resis-
tivity problem will be discussed in detail.
FL’NDA,VENTAL RELATIONS
The flow of steady electric current in a nonuni-
form medium containing a current source iS gOV-
erned by the relation
-V.
[ xx+>
,
me,7
Y, z)
1
aQ(x, Y, z)
= _____
at ’ (1)
where p is the resistivity of the medium [ohm-m],
u is the electric scalar potential [volt], Q is the
charge density [coul mm3]. We shall simplify the
notations by substituting
aQ =
--
at dx, Y9 z)
and
l/P(X, y, z) = CJ(x, y. z)
where 4 is the current density [amp m -“I and CIis
the conductivity of the medium [mhos mm’].
When the system represented by (1) is restricted
to two dimensions (x, z). we obtain
In (2), q must now be interpreted as a variable cur-
rent density in amp/m’ in a typical two-d-imen-
sional xz cross-section but per meter length in the
third direction, y. Relation (2) represents an ex-
Resistivity Modeling 63
ample of a self-adjoint elliptic system (Young,
1962). Such systems occur very frequently and can
be accurately solved by a finite-difference relaxa-
tion technique.
In the resistivity method, both the source (posi-
tive current electrode) and the sink (negative cur-
rent elec<rode) are located at the surface of the
ground: however. the following treatment is more
general and permits the location of current elec-
trodes anywhere in the medium.
resistivity PROBLElLl DISCRETIZED
For the purpose of finite-difference modeling,
the semiinfinite medium must be made finite by
introducing an artificial boundary. Such a bound-
ary is seen in Figure I, which represents a vertical
cross-section of the ground. The portion of the
boundary from A to B corresponds to the surface
of the ground, whereas the remaining boundary of
the model, BC‘DA. is fictitious. The continuous
medium can not be discretized by dividing it into
a number of rectangular cells and replacing each
cell by a point. Thus, the point P represents the
rectangular area abed. We shall call each of these
points an element of the discretired medium. In
order to emphasize the fact that the corrdinates
(x, Z) of an element can only take on integral val-
ues, the coordinates will be expressed as (i, j).
Moreover, to simplify the treatment to follow, the
origin (x = 0,~ = 0) will correspond to (i = l,j =
0).
Since the current supplied to the ground by
means of electrodes is commonly expressed in am-
peres, whereas the current density q appearing in
(2) represents the rate of charge generation (or ab-
sorption) per unit area, it would be desirable to
express q in terms of the current flowing into the
ground. Suppose the element P (Figure I) repre-
sents the location of a current electrode supplying
I amperes into the subsurface. Let us denote its
neighboring elements by E. W, N. and S (Figure
2). Since the point P represents the shaded area
abed, the current density qp due to the electrode
at P will be given by
qp = I/( Area abed)
= 41/I(h + ~,V)(h~ + h,)l. (3)
Now we consider the case
lim f/,, = 4 ~lWs& + /r,,.)l. (4)
h\-.”
 Mufti

A(O,‘J) M B
x
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

4 0
100 200

z SCALE METERS

FIG. I. A discretized model of a geologic structure with resistivities pl, 02, and p3.

Equation (4) represents the strength of the source ground instead of inside the medium--a fact
when the current electrode is located at the sur- which is well known from the analytical analysis
face of the ground. of the resistivity problem (Van Nostrand and
In the case of uniform intervals between the Cook, 1966, p, 42).
various elements, we can write Let G denote the set of elements located inside
the boundary ABCDA (Figure I ) and I’ the set lo-
h,v = h, = h, = h, = h. cated on the boundary ABCDA. If l‘, denotes the
For this case set of elements corresponding to the surface
boundary AB und 1‘, the elements along the suh-
qp = I/hY. (3a) surface boundary BCDA, we c:tn write

Also M hen /I,, = h, = h, = h and h, f h is al- 1‘ = I’, u I’,. (5)

lowed to vary such that P approaches the surface,
we get In terms of tinite-ditference modeling, we want
to evaluate the potential function L?,%,
for a set of
elements (i.;) E G U I‘,, subject to the following
lim qP = 21/h’. (4a)
h.v-0 boundary conditions:
(a) C1.J = f(s. -_) along l’,, where the func-
Equations (3a) and (4a) show that the effective tionf(.u, 2) is either given or can be computed
strength of the source is doubled Mhen a given by other means for the set of elements (i,j) E
amount of current is supplied at the surface of the I’,.

!4 /GROUND SURFACE

1 f
hw d h,

FIG. 2. An element P representing a current source in the discrete model

 Finite-Difference Resistivity Modeling 65

,_.(i
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

(i )
j‘thE
i I s

FIG. 3. An arbitrarily chosen element P of the discrete model with its four immediate neighbors. E. N.
W, and S.

(b) au/az = 0 for the elements (i,j) E I‘,.

(c) q,., = 0, except for a subset WE G U I‘,.

In resistivity surveying, qt,, # 0 only at two ele-

ments along r,, each corresponding to the loca-
tion of a current electrode.

DERIVATION OF FINITE-DIFFERENCE
EQUATIONS

Let us consider an arbitrarily chosen element

P(i,j) E G along with its four immediate neigh-
bors, E, N, W. and S (Figure 3). By making use of
(2), we shall derive a difference-approximation for
the potential u,,, at P in terms of the potentials at
its neighboring elements.
Continuing in this fashion, one can obtain the
On account of the central-difference formula
self-adjoint finite-difrerence equivalent of (2) in
(Smith, 1965. p. 6), the following approximate re-
the following form:
lation is valid at the point P.

= n.j+hs,/2 ut
v______Lti?!Y2
-
(U,,I,,,E _ v,,i>
. (6)
(hE + hw)/‘2
Therefore. for the point (i,j + hs/2) w’e can write

= hE1(vt,j+hE
- u,,j 13 (7)

au
(“ax ) LjihE/2

= - v,,j>.
u,.i+,,c,2hE’(v,.i+h~ (8)

Similarly, for the point (i. j - hw/2) we get + 4%., = 0, (11)

JV
u here y,., is pii en by (3). We can rew,rite (I I ) in an
( “ax ) <,I 11,,2
abbreviated form:

r, IFhH ,211,,!(c, , - I:,,, ,#,I. (9)

aEV!.,thB + a.X’Vt-h,v,j + alrVt,j-h,,

A second application of the central-difference for-

mula yields + ~su,+/,.Y.; - apc,,j + qz.; = 0, (12)
 66 Mufti

where where q,,, is given by (4).

QE = 2Ut,i+6*,2[hE(hE + &)I-’ Let us now consider the case of uniform inter-

Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

vals by setting
ahI = 2a ,-hN,a.i[M~S + hd-’
h, = h, = h, = h, = h. (18)
aw = 2ui,i--hR.,z[hw(hE
+ &~)I-’ (13)
By a suitable choice of scale we can set h = I. Re-
(Y&s= 20.,+hS/2,i[hS(hN + h)l-~ lation (12) now simplifies to

(YP = ffE + ffN + aT?J + a.9.

(YEUi.jlI + (Y,VV,-1.; + (Ylr’L’,.,-1 + ~.sL’,+1.;

Hitherto we have assumed that P(i, j) E G. - apc,,j + 9t.i = 0, (19)

Now let P(i,j) E rl, i.e., it is located on the sur-
face of the ground (Figure 4). Since z = 0 corre- and
sponds to the first row of elements, the coordi-
nates of the element P will be (I, j). Let us
temporarily introduce a fictitious row of elements
above the surface of the ground such that

Consequently,

Ng = NS. (15) When the element P(i, j) is located in the surface

of the ground (19) reduces to
For the elements (I, j) we must have

CYEVl ia, + 2ffsD2., + ~11-~~1.I-,

C3V

az r=O = O* - ff,‘C, ,, + q, , = 0. (2 1)
This condition will be satisfied if we set It may be recalled that when P( I, j) corresponds
to the negative current electrode, ql,, must be re-
&+h?;,,~ U,-h,~.]. (16) placed by -q,.,; when it corresponds to a poten-
In view of (15) and (16). (12) can now be replaced tial electrode. q,,, = 0.
We are now in a position to evaluate u,,, at all
by
elements (i,,;) E G U I‘, in terms of the potential
ffEVl.j+h~ + 2%V,+hs.j + ffU.Vl,j-h,,
at their four immediate neighbors. Since a,,, for (i,
j) E I’, will be supplied, we have taken care 01
the entire set of elements (i, j) E G LJ r.

N
--___--_--_--_---~------_-_
J-f--- i=O
-y-
h, 1 GROUND
SURFACE
+ :;
i = 1
W P E

hs

f :;
i=2
S
hw h, -1

FIG. 4. Neumann type boundary containing a current source P.

 Finite-Difference Resistivity Modeling 67

EVALUATION OF THE POTEhTIAL FIELD 0 1 2 j- n Il+1

1
3
Re.ri.rticit~~matrix
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

5
Let the area A BCD (Figure I ) be replaced by a
set of (nr + I) (n + 2) elements as illustrated in
Figure 5. The problem is to compute the values of
the potential u,., for (i,j)
E G IJ I’,, (i = I, 2. .,
i
nz;j = I, 2, n), when the set of values c(i,j) for
(i, ,j) E I’, is given. For solving practical prob- m

lems, both IN and n must be quite large. However,

for the purpose of understanding some of the de- m+l
tails to follow it would suffice to consider a very
small field obtained by setting m = n = 3 (Figure FIG. 5. The area ABCD of Figure I. discretized
6). In order to simplify the treatment we shall as- into a set of (m + I) X (n + 2) elements.
sume uniform grid intervals, although this treat-
ment is equally valid for nonuniform grid inter-
vals. We shall express the potential at each point 0(~(6)un - cu.s(h)o, - cu,,-(6)~; - ~c(6)r<1
in terms of the potential at its four immediate
neighbors and the set of coefficients (yE, CY.~,,No, = (IP(~) + o(E:(6)c,; = b,;, (28)
CY.~,and (yP introduced previously. But since we are
now dealing with more than one element, each as-
sociated with its ow’n four neighbors, it is conven-
ient to express the set of coefficients as (tE(k), etc.,
where (k) indicates the location of the element un-

der consideration. In Figure 6, c,,,, c,,, , cl8 are

given, and we want to compute L’,, I_+, , cgwith
the help of (19) and (21).
For element I, (21) leads to the equation

= II,. (31)
By placing the known terms on the right-hand
side, (22) can be expressed as
In matrix form the system of equations (24)
c+(l)u, - cXB(l)L’:, - 2as(l)c,
through (31) can be expressed as

= qp(l) + (Y,,.(l)CI” = b,. (23)

Similarly, for the remaining eight elements we
1 2
nave 10

%=(2)u* - (YE(2)U:< - a,r(2)c, - 20(,(2)c,

= qp(2) = b,, (24) 11 17

,~uP(~)u, - cu,v(3)a, - 2czs(3)c,;

= 9P(3) + CYE!3)C,R = b:,, (25)

12 16
aJX4)u.l - (YE(4)L.S - CYv(4)L.r - (Y.q(4)c;

= QP(4) + ~,,~(4)C,, = bd, (26)

LyP(5)US - c%(5)& - %(S)C, - (Y,J5)CJ

- %(5h = SP(5) = 65, (27) FIG. 6. A simple example of a resistivity model.

 Mufti

Relation (32) can be written in abbreviated

form
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Rv = b.

Here

R = [r,,,], i, .j = I, 2, . . 9;

v = [C,,C*, .” , U!,li
b = [b,, bZ, . , h!,]

Since the elements ri,, of the matrix R are

directly related to the distribution of resistivities
in the subsurface, R will be called the resistivity
matrix. After several simultaneous permutations
of Its rous and the corresponding columns, this
matrix can be transformed into a tridiagonal
block form. A matrix which can be reduced to this
form is said to possess property A first formulated
by Young (1954). An interesting account 01
this propert) of matrices is given by Smith (1965,
p. 82).
Another characteristic of matrix R which is of
interest to us is its directed graph. One can easily
serif) that the directed graph of this matrix is
strongly connected. Those readers who are inter-
ested in further details of directed graphs are re-
ferred to KGnig (1950).

Stahilitj- conditions

Although the above treatment involved a small

field (ttt = tt = 3), it is obviously applicable to
large fields as uell. Let us now assume that both m
and tt are quite large (Figure 5). The number of
elements for which we want to compute the field
c,,, is run. Therefore, the resistivity matrix will
have the dimensions N X N, where N = mn. The
system of equations to be solved can be expressed
3S

,F I’, ,C,,, = b,, i = 1, 2, , N, (33)

u here r,., are the elements of R. This system can

be solved by a successive overrelaxation scheme
provided the following conditions are satisfied
(Young. 1962):

r I.1 > 0, i = I, 2, ... , N. (343)

s
r, ,, 2 c lr,., 1, i = 1, 2, . , N. (34b)
,=1
Ifl
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
 70 Mufti

+ a,,(i, j)c(,‘L,‘,)l The SLOR method is based on computing the

improv,ed estimate of u,,, simultaneously along the
entire row rather than for each individual element
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

of the model separately. Imagine that the (k + I )th

iteration is in progress and we are about to start
with the ith row. For the first element along this
row’ (j = I ). we can write
where w = I. In SOR, we overrelax the field at
every point by using an acceleration parameter I
< w < 2. When dealing with nonuniform media - aAi, l)O, ,2 - a,y(i, l)c’,‘!+,‘,:
there is no easy way to determine an optimum
value of this parameter. It is commonly deter- - Lys( i ) 1)c(,‘:)l , - (I#,I = 0. (43)
mined by using the relation
In (43). C,,, denotes the elements of an auxiliary
vector

V,
- = [C,,l, i-,.2, ..’ , C,,.l,
where p(G) is given by (3X).
In practice, one starts the finite-difference com- i = I, 2, ... , m,
putations by assuming w = I and evaluates p(G) from which we shall compute an improved esti-
periodically, say, at the end of every 5th iteration. mate of the vector v, defined by (42).
After a number of iterations. the variation in the We also note that o,,, belongs to the boundary
magnitude of p(G) becomes less and less and p(G) and is known: 1;,+$:, and IJ,!:.:’ have already
tends to an asymptotic value. This value ofp(G) is been computed, during the kth and (k + I)th
used for computing the optimum value of w. Ex- iteration, respectively, and can be treated as
perience with resistivity modeling indicates that “known” information. By transferring all the
the optimum value of w is much more dependent known terms on the right-hand side, (43) can be
on the order of the resistivity matrix R, and is not expressed as
very sensitive to changes in magnitude of the ele-
ments of R. This is very fortunate because, as long ap(i, l)C,,, - crE(i, l)C,,,
as we do not change the size of the potential field
= crs(i, l)l;i’l;l: + au,(i, I)cj’:‘,,,
to be investigated, the same value of w may be
used for investigating different models for which + (Y&i, l)C,,” + [I, 1 = 11. (44)
the geology does not change drastically.
We shall now describe a more efficient pro- Similarly, for the next (n - 2) elements of the ith
cedure based on the SLOR method for computing row of the model, we can write
the potential field. In the above treatment the set
of data (u,.,} appeared as an N-dimensional vec- -(~,r~(i, .i)fi,.,-l + c.uAi, .i)t,.,
tor. Thus, for the simple model represented by
- CQ(~, j)Ci.,+l
Figure 6, both i and j ranged from 1 to 3 and N
was equal to 9. Let us now consider a series of = as(i, j)u!k:t:
vectors:
+ adi, j)u,+l,,
lk) + YT.1 = 17,
V, = [c,.1,c,.*, ..’ ,L’,,,l,
j = 2, 3, ... ) (n - 1). (45)
i = 1, 2, .. . , nr. (42)
When j = n, we have
Here m and n denote, respectively, the number of
rows and columns defining the elements of the --cr&i, n)c, .n--l + OrAi, n)fi, .n
resistivity model. They should not be confused
= a&i, n)u,,,+l + o(Ai, n>~‘~‘?~‘,‘,
with the rows and columns of the resistivity mat-
rix R. which is a square matrix of order N = mn. + cus(i, n)~:ik>‘t.” + Yt.n = 1”. (46)
The components of vt given on the right-hand side
of (42) represent the values of the potential along The system of equations (44) (45), and (46) can
the ith row of the model. be expressed as
 Finite-Difference Resistivity Modeling 71

F In compact notation, (47) is equivalent to

9
-” 2 . . .
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

i R,:, = t,, i = 1,2, ... , m. (48)

/ 1
II
Since R, is a tridiagonal matrix, the vector q1
-. n c1
,j: ,j * 3s can be readily determined by the method of Gaus-
I I sian elimination (Faddeeva, 1959, p. 65). The set
I of values obtained in this manner is further
modified by overrelaxation to obtain the (k + I )th
,^
estimate of the vector v,:
I
z -2
.,^ ^
-3
Vi”+‘I = vjk)+ WdV, - vik’),
2 ir”
I i = 1,2, ... m, (49)
h

I where w, represents an optimum value of the ac-

. E c celeration parameter when we are using the SLOR
._r; .-,^ method. The subscript 1 is used to emphasize the
YY
fact that wI is different from the parameter w
tl 21
I previously introduced in the discussion of the
SOR method: wI is given by a relation similar to
h
(41):

2
wr =
1+ [l - pi(G)] “’ ’

p,(G) is the spectral radius of the Gauss-Seidel

- . . iteration matrix. In the case of the SLOR
m
. .
4 . . method, pi(G) is determined by setting w1 = I.
FI Apart from that, the procedure for evaluating
p!(G) is exactly the same as the procedure pre-
viously described for p(G).

CONSIDERATIONS RELATED TO
MODEL DESIGN

Problems related to boundaries and grid intervals

When the area ABCD shown in Figure I is

large and the boundary BCDA is sufficiently far
from the current electrodes, we can neglect the
amount of current flowing across this boundary.
This leads to the following Neumann boundary
I
I 9 condition.
 72 Mufti

a a,
-__ z)
ax = 0, i = 1, 2, ... , m + 1; j = 0, n + 1;
Cl.,)
(50)
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

a-__
4x3 2)
az = 0, j = 0, 1, ... , n + 1; i=m+l. I
(7 ,f) i

When we are investigating less resistive media, the strip they are expanded horizontally in three large
values of the potential along the boundary BCD.4 steps. As a result, the structures enclosed by
may not be too different from zero; in that case, ABCDA now become a “local” perturbation in a
we can introduce a somewhat simpler Dirichlet much larger zone enclosed by the external bound-
boundary condition: ary of the model A’BC
’D’A
’ .’ This zone would be
suffkiently large provided we get virtually identi-
v(i, j) = 0, for all (i, j) E r,. (51)
cal results in the inner zone ABCDA. irrespective
Numerical experiments on resistivity modeling of whether we use (50) or (5 I ) along the boundary
indicate that the size of the resistivity matrix must B’CD
’A’’ for evaluating the potential field.
be extremely large before we can get accurate Optimization studies along these lines suggest
results by using either (50) or (51); otherwise, (50) that the effects of the subsurface boundaries be-
leads to an overestimation and (51) an under- come negligible when we set the intervals BE, EF,
estimation of the actual field which is only and FE’ equal. respectively, to IO, 100, and 1000
bounded at the surface of the ground. This prob- times the grid intervals used for discretizing the
lem can be overcome by replacing the subsurface area ABCD. The same argument applies to the
boundary by a strip of material of large effective horizontal portion of the termination strip, except
width, as shown in Figure 7(a). Imagine that we that the grid intervals are expanded downward.
set up a model by using uniform grid intervals f-‘or most problems. accurate results can be
everywhere, including the zone occupied by the obtained by dividing the horizontal distance AB
termination strip. Within the strip, the resistivities into 50 intervals and the vertical distance BC
PI, P2. ” are assigned as if it were a natural into 25 intervals. It may. be pointed out that a
extension of the structures under investigation. modeling system based on the self-adjoint equa-
After that, the strip is expanded outward without tion remains stable for all values of grid intervals.
modifying the resistivities. An enlarged view of a Consequently. the actual values of the grid inter-
portion of the expanded strip is shown in Figure vals may be assigned arbitrarily, depending upon
7(b). We observe that the grid intervals in the the accuracy with which the model is expected to
zone of interest are still uniform, but within the represent the subsurface geology

A' A B B’ B E F B’

(b)
(a)

D’ C’

FIG. 7. (a) Extension of a resistivity model of an arbitrary structure by a termination strip. (b) An en-
larged view of a portion of this strip.
 Finite-Diference Resistivity Modeling 73

Manipulation of coefficients line joining the electrodes, the potential at any

point (x. z) in the subsurface is given by
The efficiency of the finite-difference evaluation
of the potential field is heavily dependent upon
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

those calculations which occur repeatedly. Thus,

in the SOR evaluation of the field, the use of (40)
will form the bulk of the computations. There-
fore, the set of coefficients crE, cyw, ~y,v, as, and op
for each (i. j) must be computed only once before
the iterations begin. Further examination of (40)
will suggest that this relation should preferably be
used in the form
. /(xi 2nh, + z), (54)

where

f(x; z) = C(X, z)

+ Bs(i, ib’,:‘,., + Y,.,l, (52) L is the distance between the current electrodes; h,
is the thickness of the first layer; y = p2/p,: and I
where
is the current per unit length of the line electrodes.
Relations (53) and (54) give the potential in layers
I and 2, respectively, whereas (55) represents the
potential u(x, z) of a uniform medium with resis-
tivity p,. When y = I, both (53) and (54) are
A comparison of (40) and (52) suggests that
reduced to (55).
repeated divrsions by (~~(i, j) as implied by (40)
We computed the potential field of a structure
can be avoided by computing the coefficients
consisting of two parallel layers analytically by
P E,I,,,r etc.. in advance. Moreover, instead of stor- using (53) and (54) and also by means of the
ing the tive coefficients No, (Ye, a,+.,,as. and ap for
finite-difference scheme. The computations were
each (i,j).we now need only the four coefficients
based on the following parameter values:
PE, Bw. @.V.and ps for each (i,j), thereby reducing
the siLe of the computer memory required for resistivity of the upper layer, p, = 100 ohm-m:
storing the coefficients resistivity of the lower layer. p2 = 30 ohm-m;
thickness of the upper layer, h, = 20 length
APPLICATIONS units:
current electrode separation, L = 20 length
Several computer programs based on the treat-
units;
ment presented above were developed and used
current, I = 25 ma/unit length.
for computing the potential field and apparent
resistivity curves for various structures. Some of
A short reflection on (53) and (54) will indicate
the structures selected for this purpose possessed a that the results do not depend on the specific unit
simple geometry, and their field could also be
of length used. Thus, in Figure 8, which gives the
computed analytically. This afforded a com-
computed equipotential lines, the unit of length
parison of analytical results with the data ob-
corresponds to 20 m; but if the unit of length were
tained by the use of the corresponding finite-
30 m, the same field would represent a structure
difference models.
whose upper layer is 600 m thick and receives
current at the rate of 25 ma/(30 m) from a pair of
Two parallel layers line electrodes 600 m apart. The field shows a
The potential field inside a parallel-layer earth break in the equipotential lines all along the inter-
due to a pair of infinite-line electrodes located at face between the two layers. These data are in
the surface has been investigated by Peters and excellent agreement with the analytical results ex-
Bardeen (1930). When there are only two layers cept in the immediate vicinity of the lower bound-
and the origin corresponds to the midpoint of the ary of the model.
 Mufti

C, (-) C, (+I
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

* 200
c?z
l.lI
I-
r" 400
z
u

c 600
w
a FINITE-DIFFERENCE

800

98(
0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200

HORIZONTAL DISTANCE IN METERS

FIG. 8. The potential for a two-layer structure computed by finite-difference modeling (broken lines)
and the corresponding analytical results (continuous lines).

Structure containing a dipping interface = */2n, where n is an integer and p2 ~- 0, the

potential at any point P(r, 0). located at the sur-
The structure investigated is shown in Figure 9. face of the ground (0 = 0), is given by
The interface OE separating materials of resis-
tivity p, and p2 is assumed to be so oriented that
the direction of the strike is parallel to the direc-
tion of the line electrodes. ABCDA defines the
boundary of the tinite-difference model. When p2 + (-1)” I*>$+ + “2 (-I)”
<< pI or pz >> pl, the potential along the surface k=,

of the ground can be computed analytically (Pe-

ters and Bardeen, 1930). Let r, and r2 denote the
In rz+ r2- 2r,r cos__-
W/n)
(56)
distances of the current line-electrodes C, and C,,
r: + r’ - 2r,r cos (h/n) 1’
respectively, from the point 0, taken as the origin In the finite-difference model the values of the
of the polar coordinates (r, 0). If the dip angle is (Y potential along the boundary BCE were obtained
vl
I- A 0 c, l-1 c, (+I B
z
3 0
y GROUND SURFACE
z lo-
2
E 20- p,= 100 ohm-m
E
u 30_
,’
: 4. _ P,=O.OOOOl ohm-m

Ei I I I I I I
c3 5o D 10 20 30 40 50 60 E 80 90 100 110 120 C
HORIZONTAL DISTANCE IN ARBITRARY UNITS

FIG. 9. A structure containing a dipping interface.

 Finite-Difference Resistivity Modeling 75

with the help of (55) by assuming a uniform me- sounding station, it would be convenient to make
dium of resistivity pl; for the rest of the subsurface n an odd integer. Then the location of the elec-
boundary, viz., EDA. the potential was set equal trodes will correspond to the following elements
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

to zero. The held was computed by assuming the of the model:

following parameter values:
element (1, r) for P, ;
i
r, = 26 length units,
rz = 66 length units, Y = (II t l)f2 - l/2;
p, = 100 ohm-m,
element (1, s) for P,;
pz = 0.00001 ohm-m,
cy= 45 degrees, .y = (II + 1) ‘2 + 1;2;
f = 25 ma/unit length. I
( (57)

The results are presented in Figure IO: they are element (1, u) for C,;
virtually identical with the corresponding results
a = (I? + 1);‘2 - L’2;
obtained analytically.
element (1, b) for C,;
Evaluation of apparent resistivity curves by the
reciprocity theorem b = (n + 1):‘2 + L/2. I

A computational shortcut which renders hnite- The apparent resistivity of the ground in the
difference modeling ideally suited to the eval- case of a line-source dipole (Parasnis, 1965) is
uation of vertical-sounding apparent resistivity given by
curves for the line-source Schlumberger con-
figuration can be explained as follows. In a finite-
p,(L) = “z (In 5%)’ $ , (58)
difference model, the location of electrodes will
correspond to the surface elements (IJ) E Tl , j
where AL: is the potential difference across P, and
= I, 2, ., n. Let L denote the distance between
P, and I denotes current flowing into the ground
the current line-electrodes C, and C,, and I the
per unit length of the line-source dipole. Ideally, I
distance between the potential electrodes P, and
<< L: in that case (58) reduces to
P,. In order to simulate the Schlumberger con-
figuration, in which the point-electrodes are plan-
ted symmetrically with respect to the center of the

20 30 40 50 60 70 80 90 100 110 120 130

HORIZONTAL DISTANCE IN ARBITRARY UNITS

FIG. IO. The distribution of potential along the surface of the structure shown in Figure 9 due to a
line-source dipole C, and C,. Results of analytical and finite-difference computations are shown.
 76 Mufti

In order to keep I as small as possible, let us rent remains the same when the positions of the
choose I = 2 in the model: then the potential current source and the voltmeter are
electrodes will correspond to the elements [I, (n - interchanged. In our case we are dealing with
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

1)/2] and [I, (n + 3)/2]. Since the model allows apparent resistivity. which is nothing but a nor-
for the computation of all the values c,,,, j = I, 2. malized value of this ratio. Consequently, in the
” 1‘ n, the potential difference across these ele- first modeling experiment, if we had shifted the
ments can readily be obtained. An apparent resist- current electrodes to the elements ( 1, r) and (I, s),
ivity curve represents a set of observations ob- the potential drop across the elements (I, a) and
tained by keeping the positions of P, and P, tixed (I. b) a distance L apart would again be AD. By
and successively increasing the value of L. Imag- the same reasoning, the voltage across (I, a’) and
ine that we compute the potential held again by (I. b’) must be 1~‘. because this setting of elec-
changing the separation between C, and C, to L’. trodes results from interchanging the current and
C, and C, will now correspond to the elements (I, potential electrodes in the second experiment.
a’) and (I, 6’). in accordance with (57). and the This observation leads to an important con-
apparent resistivity will be given by clusion:
By selecting a model size sufficiently large that
it accurately represents the resistivity distribution
p,(L’) = ; (In $;+)-’ A; . (60)
over the zone of interest, the entire resistivity
curve can be computed in a single step. All we
In (60) AU’ represents the potential difference need to do is compute the potential field at all
across the elements [I. (n ~ I),/21 and [I, (n + those locations which are actually occupied by the
3)/2] due to the new position of the current elec- current point-electrodes in an expanding Schlum-
trodes. Note that we have used the same amount berger configuration and by locating the current
of current in the two modeling experiments. line-electrodes at the elements of the model which
At this stage we shall make use of the reciproc- actually correspond to the original field location
ity theorem (Jordan and Balmain, 1968, p. 347), of the potential electrodes. When great depths are
which states that if a current is applied to an involved, the size of the resistivity matrix may
isotropic physical system at one pair of terminals become too large. This problem can be avoided if
and the open-circuit voltage is measured at a sec- we use nonuniform grid intervals, taking care that
ond pair of terminals, the ratio of voltage to cur- these intervals get smaller as the current elec-

lOOr I I I lllll I I I IllIf

80 -

20 - -ANALYTICAL

o FINITE DIFFERENCE

LIZ, m
FIG. I I. Computed apparent resistivity curve for a Schlumberger type line-source array vertical
sounding of a two-layer structure. Results for analytical and finite-difference calculations are shown.
 Finite-Difference Resistivity Modeling 77

100 I I I Ill11 I I I 1

80 -
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

60 _

/
7 40- 0”
AZ
E
O_ 30- 0
/

m
Q /

20 - /

4 _-\A

1 I I I II11 I I I
10
10 20 30 40 60 80 100 200 300 500

L/2, m
FIG. 12. Computed vertical sounding apparent resistivity curve for the structure shonn in Ftgure I. A
Schlumberger type line-source array was assumed.

trodes are approached. It must also be remem- computed by assuming the Schlumberger-type
bered that the choice of the grid intervals should line-source configuration with its center located at
be such that the various subsurface interfaces of point M (Figure I ).
the structure under investigation run, as far as
(‘ONC‘LUDING REMARKS
possible, half-way between the neighboring ele-
ments of the model. This is an important consid- The resistivity method has been widely used for
eration which will become more obvious from the the exploration of groundwater. Its success in this
boundary of the hatched zone shown in Figure 2. particular area of exploration is at least partly due
Figure I I shows a computed apparent resistivity to the fact that the assumption of a parallel-layer
profile for a Schlumberger-type line-source array earth, which forms the basis of conventional pro-
on a structure consisting of two parallel layers. cedures of resistivity data interpretation, is usu-
The calculations were carried out by assuming p, ally applicable to the occurrence of groundwater
= IO ohm-m, p2 = 50 ohm-m, and h, = 40 m. The in nature. Since both gas- and oil-bearing forma-
continuous curve represents analytically com- tions possess high resistivity, this method should
puted data, and the values obtained by finite- in principle be suitable to exploration for hydro-
difference modeling are indicated by circles. carbons as well.
The cases presented so far demonstrate that, for The finite-difference modeling system presented
all practical purposes, the results obtained by fi- by the author was developed for evaluating poten-
nite-difference modeling are as accurate as those tial fields and vertical sounding apparent resist-
obtained by analytical means. ivity curves for complex geologic structures com-
monly associated with the occurrence of
Apparent resistivity curvefor a complex structure
hydrocarbons. The entire treatment was restricted
Figure 12 shows the computed apparent resis- to two dimensions and the use of a line-source
tivity curve for the geologic structure shown in dipole. Some of the structures investigated pos-
Figure I. The curve corresponds to p1 = I5 ohm- sessed a simple geometry, and they could also be
m, pZ = 3 ohm-m, and p3 = 75 ohm-m; it was treated analytically. In such cases the results ob-
 78 Mufti

tained by finite-difference modeling were virtually REFERENCES

identical to the corresponding values computed Aiken. C. L.. Hastings. D. A.. and Sturpul. J. R.. 1973,
Ph! sical and computer modeling of Induced polari7a-
analytically. This suggests that finite-difTerence
tlon: Geophls. Prosp.. v. 21, p. 763-7X2.
Downloaded 12/25/14 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

modeling is 3 very powerful tool capable ol Brigham. E. 0.. and Morro\+. R. E.. 1967. The fast
yielding accurate results for ;I large variety Fourier transform: IEEE Spectrum. v. 4, p. 63-70.
baddeeva, V. N.. 19SY. Computational methods of
of two-dimensional geologic structures. The linear algebra. Neu York. Dover Publications. Inc..
procedure outlined above can be extended to ac- Translated from Russian bv Curtis D. Benster.
commodate three-dimensional structures and a Jepsen. A. F.. lY6Y. resistivity and induced polariration
modeling: Ph.D. diasertaiion. Universit) ot Cali-
point-source dipole. Unfortunately, the use 01 fornla. Berkeley.
three-dimensional finite-difference models implies Jordan. t. C.. and Balmain. K. G.. 1968,
Electromagnetic Haves and radiating systems: Engle-
prohibitive computer costs. We hope that-this
nood Cliffs. Prentice-Hall. Inc.
apparent drawback will dwindle with the rapid11 Kcinig. D.. 19.50. Theorie der endlichen und unendlichen
growing use of minicomputers. In the opinion of Graphen: Ne\\ York. Chelsea Publishing Co.
Parasni?. D. S.. 1965. Theory and practice of electric
the author, however, a more practical approach
potential and resistivity prospecting using linear cur-
would be to transform line-source Dotential lields rent electrodes: Geoexnloration. v. 3. D. 3-69.
computed for two-dimensional structures into the Peters. L. J.. and Bardeen. J.. 1930. The solution of
some theoretical problems which arise in electrlcal
corresponding point-source fields along the lines methods of geophlsicul exploration: University of
described by Tranter (1956). The computational Mlsconsln Eng. Eupt. St. Bull. 71.
Smith. G. D.. 1965, Numerical solution of partial dither-
effort required for such a transformation can be
entlal equations: Neu York. Oxford Unrversi!) Press.
greatly reduced by the use of the fast Fourier South\\ell, R. V.. 1946. Relaxation methods In theo-
transform (Brigham and Morrow, 1967). retical physics. \. I: London. Oxford Universit)
Press.
Tranter. C. J.. 1956. lntreral c
transforms in Mathe-
matlcal physics: Nem York. John M ilel & Sons. Inc.
.ACKNOWLEDGMENTS
Van Nostrand. R. G.. and Cook, K. L.. 1966. Inter-
The author is indebted to Dr. W. G. Clement, pretntion of resistlvltj data: USGS Prof. paper 449,
Washington. D.C.. U.S Govt. Prlntlng office
Dr. Rup K. Kaul. and Dr. Sven Treitel for sugges- Varfa. R. S.. 1962, Matrix iterative analysis: Engleuood
tions and criticisms; to Dr. S. N. Domenico and ClilTs. Prentice-Hall. Inc.
Young. D.. lY62. The numerical solution of elliptic and
Dr. A. C. Reynolds for reviewing the manuscript:
parabolic partial dilrerential equations. in Survey of
and to Rida Bakbak for sharing a significant bur- numerical analysis: New York. McGram-Hill Book
den of programming. Permission from Amoco co.. Inc.. p. 380.338.
Young. D.. 19.54. lteratlve methods for solving partial
Production Co. to publish this work is greatly
dilrerence equations of elliptic type: Trans. Am.
appreciated. Math. Sot.. v. 76. p. 92-l I I.