Geophysical Prospecting, 1997, 45, 809–818

Complex polynomials for the computation of 2D

gravity anomalies1
Sergio E. Oliva2 and Claudia L. Ravazzoli2

Abstract
We present an efficient algorithm using a complex variables formulation for the
computation of the gravity effect of 2D polygonal bodies having densities varying both
laterally and with depth. The first derivatives of the gravity effect are also provided in
order to enable the computation of the Jacobian matrix, which is necessary for linear
inverse gravity problems. A geophysical example based on numerical assumptions
about the density contrast on a well-studied basin area shows the applicability of the
algorithm.

Introduction
A 2D model in which the density contrast is assumed to be constant is usually the first
step in the interpretation of the anomalous gravity response due to different geological
structures in a prospect area.
In the case of sedimentary basins, the structural history and settling conditions
of deposition of the sediments as well as various diagenetic processes (e.g. re-
crystallization, cementation, compaction) influence the density of the rocks. The
correlation between the depth of burial and the density of sediments has been noticed
early by earth scientists (e.g. Athy 1930). Nevertheless, a density distribution which is
only depth dependent may not be sufficient to account for the gravity anomalies
observed in a basin (Cordell 1973), so that lateral changes in the density may be
inferred. Unfortunately, analytical solutions for the gravity field due to density
distributions which vary laterally and with depth are extremely difficult, if not
impossible, to obtain in any surveyed area. Since polynomials may accurately fit the
density distribution of many anomalous bodies and the gravity response obtained may
resemble the actual response (Rao 1986), an algorithm for the computation of the
gravity anomaly due to an arbitrary polynomial density contrast could lead to more
realistic models. In fact, several authors have presented algorithms in both the spatial
and frequency domains which partially achieve that goal (e.g. Murty and Rao 1979;
Murty, Rao and Ramakrishna 1989; Pan 1989; Guspı́ 1990).

1
Received May 1996, revision accepted March 1997.
2
Facultad de Ciencias Astronómicas y Geofı́sicas (Observatorio Astronómico), Universidad Nacional
de la Plata, Paseo del Bosque S/N, (1900) La Plata, Argentina.

q 1997 European Association of Geoscientists & Engineers 809

810 S.E. Oliva and C.L. Ravazzoli

Soon after the first algorithms for 2D polygonal bodies having an irregular cross-
section were published (Talwani, Worzel and Landisman 1959), Jung (1960) proposed
the use of complex variables for the expression of the gravity potential of polygonal
bodies. Many years later, Kwok (1989) transformed double integrals into conjugate
complex contour integrals in order to shorten the analytical calculations for obtaining
the gravity effect. He elegantly obtained equations which even included polynomial
density functions. In this paper the same idea is applied to the general case of a 2D
polygonal body with laterally varying and depth-dependent density contrast given
by an arbitrary polynomial law. We have developed an analytical solution for the
computation of the gravity effect and also the first-order partial derivatives, aimed at
a subsequent linear gravity inversion by standard algorithms.
In order to prove the applicability of our method, we include a numerical example
in which high-order polynomials give an adequate fit to a known density contrast for
a fault model.

Description of the method

Let Q be the cross-section of a closed 2D body, with a density contrast given by a real
bounded function rðx; zÞ. Then at a point ðx0 ; ¹hÞ, h $ 0, the gravity effect is given by
…… zþh
gðx0 ; ¹hÞ ¼ 2G rðx; zÞ 2 2
dxdz; ð1Þ
Q ðx ¹ x 0 þ ðz þ hÞ
Þ
where G is Newton’s gravitational constant and the variable z is taken positive
downwards. As is well known, the boundedness of r ensures the convergence of the
parametric integral (1) everywhere.
Next we will assume that at every point in the body the density r can be fitted by an
mth degree polynomial of the form
X
m X
n
rðx; zÞ ¼ an¹s;s xn¹s zs ; n $ s; ð2Þ
n¼0 s¼0
where the coefficients an¹s;s are real constants. This polynomial may be the result of
either a fitting procedure or a Taylor expansion of a known analytical density function
for the body. By means of the complex substitution,
p
w ¼ x þ iz; i¼ ¹1;
w̄ ¼ x ¹ iz; ð3Þ
P0 ¼ x0 ¹ ih;
we can rewrite (1) and (2) in the complex plane. From (3) and using a binomial
expansion, the real density function rðx; zÞ can be written as
Xm X n
an¹s;s s n¹s
rðw; w̄Þ ¼ n i ðw̄ þ wÞ ðw̄ ¹ wÞs
n¼0 s¼0
2
X
m X
n
an¹s;s X
n
¼ is ðn ¹ sÞ!s! cp w̄n¹p w p ; ð4Þ
n¼0 s¼0
2n p¼0

q 1997 European Association of Geoscientists & Engineers, Geophysical Prospecting, 45, 809–818
 Computation gravity anomalies 811

where rðw; w̄Þ : C → R, and cp is a real coefficient given by,

8Pp
< j ¼0 g pj; p # ðn ¹ sÞ;
cp ¼ P
for p < s;
: p¹1
j ¼0 g pj; p > ðn ¹ sÞ;

and
8Pp
< j ¼ p¹s g p j ; p # ðn ¹ sÞ;
for p $ s; cp ¼ P
: n¹s j ¼ p¹s g pj; p > ðn ¹ sÞ;
where
ð¹1Þ p¹j
g pj ¼ :
ðn ¹ s ¹ j Þ! j !ðs ¹ p þ j Þ!ð p ¹ j Þ!
The symbols C and R denote the domains of the complex and real numbers,
respectively. After rewriting part of the integrand in (1) as in Kwok (1989), with the
notation given in (3), it follows that
 1
 zþh
¹Im ¼ ð5Þ
ðx ¹ x0 Þ2 þ ðz þ hÞ2
;
w ¹ P0
so that the gravity effect (1) at any gravity station P0 can now be rewritten in terms of
the conjugate complex variables w and w̄ in the form
……  
1
gðP0 Þ ¼ ¹ 2G rðq; w̄Þ Im dxdz
w ¹ P0
… … 
Q

rðw; w̄Þ
¼ ¹ 2G Im dxdz ; ð6Þ
Q w ¹ P0

which is a modified expression of that given by Jung (1960). Next we recall the
complex form of Green’s theorem in a plane, given by
‡ …… ∂B
Bðw; w̄Þdw ¼ 2i dxdz; ð7Þ
G Q ∂w̄
where Bðw; w̄Þ is a continuous function with continuous partial derivatives within the
region Q and on its border G (Spiegel 1964). Then using (4), (6) and (7), for w Þ P0
we get
1 X m X
n
an¹s;s X
n
cp
Bðw; w̄Þ ¼ is ðn ¹ sÞ!s! w̄l w p ; where l ¼ n ¹ p þ 1:
w ¹ P0 n ¼ 0 s¼0
2n p¼0
l

Thus, (6) becomes

(‡ )
X
m X
n
an¹s;s X
n
cp w̄l w p
gðP0 Þ ¼ 2G Re s
i ðn ¹ sÞ!s! dw ; P0 Ó G: ð8Þ
G n¼0 s¼0 2nþ1 p¼0
l ðw ¹ P0 Þ

q 1997 European Association of Geoscientists & Engineers, Geophysical Prospecting, 45, 809–818
812 S.E. Oliva and C.L. Ravazzoli

Figure 1. Representation of a 2D polygonal body in the ðw; w̄Þ-plane. P0 is a station located at

ðx0 ; ¹hÞ and wk denotes the position of a vertex at ðxk ; zk Þ.

Now we consider G to be a closed N-sided polygon with vertices wk looped in a

counter-clockwise sense (Fig. 1). Then the line integral in (8) can be expressed as the
sum of N definite integrals. First note that over the kth side the following relationship
between w and w̄ holds (Kwok 1989):
w̄ ¼ ak w þ bk ;
where
Dw̄k 2iSk
ak ¼ ; bk ¼ ; k ¼ 1; 2; …N; ð9Þ
Dwk Dwk
and
Dwk ¼ wkþ1 ¹ wk ;
Dw̄k ¼ w̄kþ1 ¹ w̄k ; ð10Þ
Sk ¼ xk zkþ1 ¹ xkþ1 zk :
The constant coefficients ak, bk hold for the kth side and DwN ¼ w1 ¹ wN for the Nth
side. By means of the binomial expansion, the w̄l term can be written as
X
l  j
1 ak
l
w̄ ¼ l !b nþ1
k bk
¹p
w j; for bk Þ 0; ð11Þ
j ¼0
ðl ¹ j Þ! j ! bk

so we can therefore write the gravity effect (8) as a sum over the sides of the polygonal
boundary
X
N
gðP0 Þ ¼ gk ðP0 Þ; ð12Þ
k¼1

q 1997 European Association of Geoscientists & Engineers, Geophysical Prospecting, 45, 809–818
 Computation gravity anomalies 813

with
(
X
m
b nþ1 X
n X
n
gk ðP0 Þ ¼ 2G Re k
an¹s;s is ðn ¹ sÞ!s! cp ðn ¹ pÞ!b ¹p
k
n¼0
2nþ1 s¼0 p¼0
 j … w )
X
l
1 ak kþ1 w j þp
× dw ; 1 # k # N: ð13Þ
j ¼0
ðl ¹ j Þ! j ! bk wk w ¹ P0

The polynomial quotient in the above integral can be written as

w j þp X
j þ p¹1
P j0þ p
¼ P0t w j þ p¹1¹t þ ; ð14Þ
w ¹ P0 t¼0
w ¹ P0

so that after substituting (9) and (14) in (13), we obtain

(  nþ1 X
X
m
nþ1 Sk n
gk ðP0 Þ ¼ 2G Re ðiDw̄k Þ an¹s;s is ðn ¹ sÞ!s!
n¼0
jDwk j2 s¼0

X
n  p X
l  j
¹iDwk 1 ¹iDwk
× cp ðn ¹ pÞ!
p¼0
2Sk j ¼0
ðl ¹ j Þ! j ! 2Sk
"j þ p¹1  #)
X P0t w ¹ P0
× ðw jkþ1
þ p¹t
¹ w jkþ p¹t Þ þ P j0þ p ln kþ1 ;
t¼0
jþp¹t wk ¹ P 0

for P0 Þ wk ; 1 # k # N; ð15Þ
where only the principal value of the logarithm should be taken. It should be noted
that, as the vertices are looped counter-clockwise, when k ¼ N the following vertex
(i.e. k þ 1) will be number one.
In the particular case when the kth side is aligned with the origin, Sk ¼ 0 and then
Dwk w̄ ¼ Dw̄k w. By rearranging the constant factors of (15) in the j-index summation
and after a limit analysis when Sk → 0, it can be shown that the only non-zero term
occurs when j ¼ l.
Equation (15) is completely general under the assumptions previously mentioned
and it is, at the same time, computationally efficient because the direct computation
of several complex factors can be skipped due to the recurrence of higher powers of
the imaginary unit i, and also because of the fact that one term of the innermost
summation is actually the difference between two complex polynomials, which is
generally not a time-consuming operation.

Partial derivatives
Next we will give expressions for the first-order derivatives of gk with respect to the
vertex coordinates wk and wkþ1 which are necessary for the linear inverse gravity
problem.

q 1997 European Association of Geoscientists & Engineers, Geophysical Prospecting, 45, 809–818
814 S.E. Oliva and C.L. Ravazzoli

Let us introduce an index q which takes the value k or k þ 1. Rearranging (15) and
after a little algebra, it can be shown that
( m   n ¹iDw p
∂gk X iSk Dw̄k n X X
n
k
ðP Þ ¼ 2G Re an¹s;s is ðn ¹ sÞ!s! cp ðn ¹ pÞ!
∂wq 0 n¼0
jDwk j2 s¼0 p¼0
2Sk
 j )
X
l
1 ¹iDw̄k
× DjðqÞ ; for q ¼ k or q ¼ k þ 1; ð16Þ
j ¼0
ðl ¹ j Þ! j ! 2Sk

where, for q ¼ k,
( 
¹Dw̄k Dw̄k
DðkÞ
j ¼ ðn ¹ p ¹ j þ 1Þw̄kþ1 þ iSk ðn ¹ p þ 1 ¹ 2j Þ
2jDwk j2 jDwk j2
" j þ p ¹1  #
X P0t w ¹ P0
× ðw jkþ1
þ p ¹t
¹ w jkþ p¹t Þ þ P j0þ p ln kþ1 þ
t¼0
jþp¹t wk ¹ P0
" j þ p ¹1 #)
X P j0þ p
¹ iSk P0t w jkþ p ¹t¹1 þ ; ð17Þ
t¼0
wk ¹ P 0

and for q ¼ k þ 1,
( 
Dw̄k Dw̄k
Dðkþ1Þ
j ¼ ðn ¹ p ¹ j þ 1Þw̄k ¹ iSk ðn ¹ p þ 1Þ
2jDwk j2 jDwk j2
" j þ p ¹1  #
X P0t w ¹ P0
× ðw jkþ1
þ p ¹t
¹ w jkþ p ¹t Þ þ P j0þ p ln kþ1 þ
t¼0
jþp¹t wk ¹ P0
" j þ p ¹1 #)
X P j0þ p
þ iSk P0t w jkþ1
þ p ¹t¹1
þ : ð18Þ
t¼0
wkþ1 ¹ P0

Equations (16), (17) and (18) lead to the partial derivatives of gk with respect to the
vertices k and k þ 1 in the complex plane. Thus all the elements of the Jacobian matrix
can be obtained promptly because, in the case of an automatic inversion, several of the
factors and terms of the elements could have been stored during the computation of
the gravity anomaly for that iteration. With respect to the convergence of the iterative
process in the complex plane, special attention must be paid to every multivalued
complex function at the time of programming an application software. For instance,
sign variations around a null imaginary part of a logarithm bracket due to computer
round-off errors might give an argument out-of-phase by p, which would finally yield
oscillations or even divergence from a solution. Nevertheless, appropriate controls in
the source program for the true signs of the real and imaginary parts of those
functions could readily overcome the problem.

q 1997 European Association of Geoscientists & Engineers, Geophysical Prospecting, 45, 809–818
 Computation gravity anomalies 815

Figure 2. Polygonal model of the sedimentary basin adjacent to the master fault associated with
the Rio Grande Rift in New Mexico (after Pan 1989). The origin x ¼ 0 is taken at the fault
outcrop.

Geophysical example
As an application of the method described above, we will consider the example given
by Pan (1989), who computed the gravity effect of the sedimentary basin adjacent to
the master fault associated with the Rio Grande rift in New Mexico (Fig. 2). This
problem was first considered by Cordell (1979) who tentatively proposed a sedi-
mentary model with a gradational decrease of density from the fault basinwards over
a distance of about 5 km and up to a depth of 3 km.
Although some simple density laws, such as constant, linear and exponential, are
suitable for sedimentary basin models (e.g. Cordell 1973; Murthy and Rao 1979;
Granser 1987; Guspı́ 1990), in certain cases it may occur that more complicated
functions are necessary in order to model satisfactorily in geophysical terms, the
geology of the prospect target. Since the use of analytical expressions in the space
domain may lead to unsurmountable mathematical complications (Murthy and Rao
1979), the use of polynomial approximations is clearly convenient.
For these reasons, the main objective of this section is to test the method in the area
of interest, with different approximated polynomial densities. We would like to
emphasize that our purpose at this time is not to propose a new density model for the
area.
By means of the least-squares method, we fitted several increasingly higher real
two-variable polynomials to a set of values inferred from the piecewise linear
smoothed density distribution given by Pan (1989). These values were uniformly
distributed throughout the region occupied by the model (Fig. 2) with null density
contrast values assigned everywhere else.
In order to choose the best approximation for the density distribution, we
computed, for each polynomial and its coefficients, classical statistical tests, such
as standard deviation, Student t-tests, correlation coefficients, etc., as described by
Davis (1973). From this analysis we concluded that at least a fourth-order polynomial
was needed for appropriate representation of the density contrast data that had been
numerically generated. In Fig. 3 we show a fifth-degree polynomial, where some
positive oscillations (in the region outside the model) were excluded in order to
improve the appearance.

q 1997 European Association of Geoscientists & Engineers, Geophysical Prospecting, 45, 809–818
816 S.E. Oliva and C.L. Ravazzoli

Figure 3. Density contrast obtained by adjusting a 5th degree polynomial to the smoothed
piecewise linear function given by Pan (1989).

Next we sampled graphically the field gravity profile associated with the true
geological structure (Pan 1989) and we later computed the gravity anomalies with
three different polynomial density contrasts, as shown in Fig. 4.
All of the three polynomials fitted the original data in the same way statistically, and
their respective gravity anomalies produced a rms error of less than 61.6 mGal.

Figure 4. Observed gravity profile across the master fault of the Rio Grande rift (after Pan
1989) and computed anomalies for different polynomial density contrasts. No rms error is
beyond the 61.6 mGal range.

q 1997 European Association of Geoscientists & Engineers, Geophysical Prospecting, 45, 809–818
 Computation gravity anomalies 817

Higher degrees than the sixth produced marked density contrast oscillations which
led to much greater local errors.
Note that the greatest divergence of all the computed gravity responses from the
expected values occur near the abscissae x ¼ 0 and x ¼ 5 km, where the linearized
density contrast function given by Pan (1989) shows discontinuities in depth. Other
sources of error are due to the way we sampled the gravity data and also to the
numerical fitting of the density contrast. However, we realize that the performance of
the procedure, with additional constraints and improvements on the parameters,
would yield a better adjustment and a geologically reasonable model as well.

Conclusions
Computationally efficient formulae in the complex plane are generalized in order to
evaluate the gravity effect of 2D bodies whose densities vary both laterally and with
depth and are approximated by polynomial functions. Partial derivatives were also
provided for least-squares inversion algorithms.
As the order of the density polynomial is not only restricted to constant (zero-
order), linear or some fixed integer, a wide range of models may be developed, only
limited by other kinds of constraints. Therefore with this method any density
distribution may be treated as a whole, without artificial segmentations due to
mathematical limitations of the algorithm.
Since there is plentiful geological evidence which justifies the use of complicated
mathematical functions for the accurate modelling of density contrasts, and since
polynomials may constitute suitable approximations that consequently facilitate cal-
culations, we consider the method as a possible alternative approach to the problem of
computing the gravity effect of a model with varying density, both laterally and with
depth.
Moreover, as shown by Guspı́ (1990), the polynomial order is not critical for the
convergence of the inverse problem. Thus the problem is to determine the optimal
order so as to reach an acceptable solution promptly and with the least effort.
Although some attempts have been made for the case of a depth-varying exponential
law (Guspı́ 1990), we realize that more research is needed, especially in the situation
of lateral density changes.

Acknowledgements
We thank Prof. Fernando Guspı́ for his useful suggestions for the first version of the
manuscript. We also appreciate the work of the reviewers whose remarks improved
this paper.

References
Athy L.F. 1930. Density, porosity and compaction of sedimentary rocks. Bulletin of the American
Association of Petroleum Geologists 14, 1–24.

q 1997 European Association of Geoscientists & Engineers, Geophysical Prospecting, 45, 809–818
818 S.E. Oliva and C.L. Ravazzoli

Cordell L. 1973. Gravity analysis using an exponential density-depth function—San Jacinto

graben, California. Geophysics 38, 684–690.
Cordell L. 1979. Sedimentary facies and gravity anomaly across master faults of the Rio
Grande rift in New Mexico. Geology 7, 201–205.
Davis J.C. 1973. Statistics and Data Analysis in Geology. John Wiley and Sons, Inc.
Granser H. 1987. Three-dimensional interpretation of gravity data from sedimentary basins
using an exponential density–depth function. Geophysical Prospecting 35, 1030–1041.
Guspı́ F. 1990. General 2D gravity inversion with density contrast varying with depth.
Geoexploration 26, 253–265.
Jung K. 1960. Schwerkraftverfahren in der Angewandten Geophysik. Geest & Portig. Chapter
6.521, 199–204.
Kwok Y.K. 1989. Conjugate complex variables method for the computation of gravity
anomalies. Geophysics 54, 1621–1637.
Murthy Y.V.R. and Rao D.B. 1979. Gravity anomalies of two-dimensional bodies of irregular
cross-section with density contrast varying with depth. Geophysics 44, 1525–1530.
Murthy Y.V.R., Rao D.B. and Ramakrishna P. 1989. Gravity anomalies of three-dimensional
bodies with a variable density contrast. Pure and Applied Geophysics 130, 711–719.
Pan J. 1989. Gravity anomalies of irregularly shaped 2D bodies with constant horizontal
density gradient. Geophysics 54, 528–530.
Rao D.B. 1986. Modelling of sedimentary basins from gravity anomalies with variable
density contrast. Geophysical Journal of the Royal Astronomical Society 84, 207–212.
Spiegel M. 1964. Theory and Problems of Complex Variables with an Introduction to Conformal
Mapping and its Applications. Schaum’s Outline Series, Schaum Publishing Co.
Talwani M., Worzel J.L. and Landisman M. 1959. Rapid gravity computations for two-
dimensional bodies with application to the Mendocino submarine fracture zone. Journal of
Geophysical Research 64, 49–59.

q 1997 European Association of Geoscientists & Engineers, Geophysical Prospecting, 45, 809–818