Sie sind auf Seite 1von 13

Home Search Collections Journals About Contact us My IOPscience

Complex plane integration in the modelling of electromagnetic fields in layered media: part 1.

Application to a very large loop

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2014 J. Geophys. Eng. 11 015004

(http://iopscience.iop.org/1742-2140/11/1/015004)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 200.239.65.198
This content was downloaded on 10/02/2014 at 16:51

Please note that terms and conditions apply.


Journal of Geophysics and Engineering
J. Geophys. Eng. 11 (2014) 015004 (12pp) doi:10.1088/1742-2132/11/1/015004

Complex plane integration in the modelling


of electromagnetic fields in layered media:
part 1. Application to a very large loop
Valdelı́rio da Silva e Silva 1,2 , Cı́cero Régis 1,2 and Allen Q Howard Jr 1,2
1
Programa de Pós-Graduação em Geofı́sica, Universidade Federal do Pará, R Augusto Correa,
01-Guamá. CEP 66075-110, Belém, PA, Brazil
2
National Institute of Science and Technology of Petroleum Geophysics, Universidade Federal da Bahia,
Instituto de Geociências, sala 312-C, Campus de Ondina, 40170-115, Salvador, Brazil
E-mail: valdel@ufpa.br, terragraf@gmail.com and ciceroregis@gmail.com

Received 28 June 2013, revised 8 January 2014


Accepted for publication 8 January 2014
Published 31 January 2014

Abstract
This paper analyses the details of a procedure for the numerical integration of Hankel
transforms in the calculation of the electromagnetic fields generated by a large horizontal loop
over a 1D earth. The method performs the integration by deforming the integration path into
the complex plane and applying Cauchy’s theorem on a modified version of the integrand.
The modification is the replacement of the Bessel functions J0 and J1 by the Hankel functions
H0(1) and H1(1) respectively. The integration in the complex plane takes advantage of the
exponentially decaying behaviour of the Hankel functions, allowing calculation on very small
segments, instead of the infinite line of the original improper integrals. A crucial point in this
problem is the location of the poles. The companion paper shows two methods to estimate the
pole locations. We have used this method to calculate the fields of very large loops. Our results
show that this method allows the estimation of the integrals with fewer evaluations of the
integrand functions than other methods.

Keywords: numerical integration, Hankel transform, large loop


(Some figures may appear in colour only in the online journal)

1. Introduction k2 = −iωμσ is the frequency dependent complex wave


number associated with the physical properties of the half-
When working with geophysical electromagnetic problems space, in quasi-static conditions, a is the loop radius and r
one frequently faces the problem of calculating integrals of is the distance from the centre of the loop. This expression,
the kind: as all the calculations presented in this paper, uses the time
 ∞
factor e+iωt .
f (x) = F (λ) K(λ x) dλ, (1)
0 For large values of r and a, this integrand will be a strongly
where the kernel K of the integral is an oscillating function, oscillating function of λ. This is a particular case of the kind
usually a trigonometric or Bessel function. One simple of integral that appears when we calculate the loop fields in
example is the integral a layered earth. In the general case, the integrand function
 ∞  depends on the characteristics of every layer and it does not
1
I(r) = √ √ J1 (λ a)J1 (λ r)λ dλ, (2) have a simple expression like that of equation (2).
0 λ2 + λ2 − k 2 Such integrals present a difficult challenge for the usual
which is part of the frequency domain solution for the numerical methods. The traditional method to evaluate them
electric field generated by a horizontal loop of current over is to apply digital filters, like in the so-called fast Hankel
a homogeneous half-space at a point on the surface. Here, transform (Anderson 1979, Christensen 1990, Guptasarma and

1742-2132/14/015004+12$33.00 © 2014 Sinopec Geophysical Research Institute Printed in the UK 1


J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

Im λ

pa

3
th

th
pa
1
path 2 Re λ

Figure 1. Integration path in the complex plane.

Singh 1997). However, there are important cases in which actually more computationally efficient to calculate the fields
the filters will generate inaccurate answers, especially when from the circular loop, considering that they are a good
products of Bessel functions are part of the integrand, as in approximation to the fields from the square loop at distances
the case of equation (2). An alternative is to use asymptotic greater than three times the loop radius (Machado 2009, in
approximations to the integrand functions for large values Portuguese).
of λ. This is useful in the cases when the integral with the Taking the integration paths to the complex plane is an old
asymptotic function has an analytical solution (Ryu et al idea. It has been applied by Cauchy himself to integrals with
1970, Anderson 1979, Fullagar and Oldenburg 1984, Singh trigonometric kernels. The use in Fourier–Bessel transforms
and Mogi 2005), which is far from being the general case. has been suggested by several authors (Anderson et al 1986,
Another approach uses a continued fraction representation of Howard 2000), although not in quite the same way that is
the sum of integrations between the zeros of the Bessel function presented in this paper, where the paths in the complex plane
(Chave 1983). Recently, a quick solution was presented by Key are simply straight lines at 45◦ from the real axis, which are
(2012), using a quadrature in sub-intervals of the integration simple and effective paths.
domain and implementing the epsilon algorithm (Davis and
By using contour integration in the complex plane, it is
Rabinowitz 1975) as an extrapolation scheme to accelerate
possible to calculate the integrals accurately over a wide range
convergence, a method named quadrature with extrapolation—
of frequencies and measurement positions, on the surface as
QWE.
well as at any depth inside the layered earth. Our results show
This paper presents a detailed analysis of the application
that this method allows for the calculation of these integrals
of the method of performing the integration by a suitable
transformation of the integrand to a function of a complex with considerably fewer evaluations of the integrand functions
variable, deforming the path of integration into the than the other methods, particularly in comparison with the
complex plane, and applying Cauchy’s integral theorem. The very efficient QWE algorithm.
transformation changes the kernel function into one that
will vanish rapidly in the complex plane, so that we can 2. The method
integrate over small intervals, instead of over the infinite one
of the original problem. This procedure is valid generally and We seek to transform the original oscillatory integrand
yields results whose accuracy depends only on the numerical function into one that decreases rapidly in the complex plane
integration over finite short intervals and on the right choice with increasing |λ|, while still yielding the same result for the
of intervals.
complete integral. Such transformations take advantage of the
The motivation for this work is the modelling of data for
behaviour of the new function by deforming the integration
the electromagnetic multi-frequency method currently under
path into the complex plane, as shown in figure 1. In the
development in Brazil (Dias et al 2005, 2006). The source used
region between path (1) and the real axis the new function
in this geophysical method is a loop of hundreds of meters
will be analytical, so that Cauchy’s integral theorem applies
in radius located on the surface. In the modelling of data
for this method, the integrals involve extremely oscillatory and integration on a closed path in this region will give a null
functions. So far, the method has used a square loop as a result. Consider the closed path linking the real line to the one
source, because it is the easiest configuration to set on the at path (1) at an infinite distance from the point −α: since the
field. However, for modelling purposes, the calculation of the new function vanishes as |λ| increases, the integral from an
fields of a rectangular loop involves either an area integration infinitely distant point on the path (1) to the point −α will
of the responses of several vertical magnetic dipoles(VMDs) be equal to the integral on the real axis from −∞ to −α. The
(Ward and Hohmann 1987, p 217) or the line integration of the same is true for the region on the right: the integrals from point
responses of horizontal electrical dipoles or of the electric α to infinity following either the real axis or path (3) will yield
field of a dipole source (Poddar 1983) around the circuit the same result. Now, since the modified function becomes
of the loop. All of these options require multiple numerical vanishingly small at short distances from points −α and α on
evaluations of an oscillatory integral to compose the final paths (1) and (3), three small segments replace the infinite real
response. However, modelling the circular loop requires only path and the numerical integration is performed on these short
one numerical integration of the Hankel transform, so it is segments instead of the original improper integral.

2
J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
−0.2 −0.2
−0.4 −0.4
−0.6 Re{H(1)} −0.6 Re{H(1)}
o 1
(1) (1)
−0.8 Im{Ho } −0.8 Im{H1 }
−1 J0 −1 J1

−10 −8 −6 −4 −2 0 2 4 6 8 10 −10 −8 −6 −4 −2 0 2 4 6 8 10
(a) (b)
(1)
Figure 2. Bessel functions of the first kind (J0 and J1 ) and of the third kind (H0 and H1(1) ).

Here we analyse two different kinds of integrals involving In which the integrals on the right are understood as
Bessel functions:  +∞  −
 ∞ (1)
F (λ)H1 (λ r) dλ = lim F (λ)H1(1) (λ r) dλ
F (λ)J1 (λ r) dλ, (3) −∞ →0
 −∞
0
 +∞
(1)
 ∞ + F (λ)H1 (λ r) dλ , (9)

G(λ)J0 (λ r) dλ, (4)
0  +∞  −
where F (λ) is an even function and G(λ) is an odd function G(λ)H0(1) (λ r) dλ
= lim G(λ)H0(1) (λ r) dλ
−∞ →0 −∞
of λ. Integrals 3 and 4 are used in the calculation of  +∞ 
(1)
the components of the electromagnetic field generated in a + G(λ)H0 (λ r) dλ , (10)
stratified earth by a horizontal circular loop on the surface. 
The functions F and G carry the information about the model for positive .
of the earth and the source (Ward and Hohmann 1987); r H0(1) and H1(1) are holomorphic functions of complex λ,
represents the horizontal distance from the source. cut along the negative real axis. They both tend to zero for
The integration method makes use of the following facts |λ| → ∞ in the sector Im{λ} > 0 (figure 3), decreasing faster
about Bessel functions (Abramowitz and Stegun 1965, Watson as r increases in their arguments.
1966): In the application to the horizontal loop problem, Hn(1) (λ)
For real λ, the Bessel functions of the first kind J1 and is multiplied by exponentials, Bessel functions of the first kind,
J0 are real. J0 is even and J1 is odd. For complex λ, both are and powers of λ, the product of which will diverge for large
entire functions, with no branch points. Both diverge for large |λ| at a much lower rate than the Hn(1) (λ)’s will tend to zero.
values of |Im{λ}|. Therefore, substitution of Hn(1) for Jn yields integrands that
According to the definitions of the Bessel functions of the will decrease quickly for Im{λ} > 0.
third kind H0(1) and H1(1) , for real λ: Now we have to find the singularities of the integrands
and the regions where they are analytical functions.
Re{Hn(1) (λ)} = Jn (λ) (λ > 0), (5)
In the study of the propagation, or diffusion, of the
electromagnetic field in conductive media, which are the norm
Re{Hn(1) (λ)} = −Jn (λ) (λ < 0). (6) in geophysical problems, the functions represented by F (λ)

In this case, Im{H0(1) (λ)} is even and Re{H0(1) (λ)} is odd, while and G(λ) will involve the functions u j = λ2 − k2j (where
Im{H1(1) (λ)} is odd and Re{H1(1) (λ)} is even. Both H0(1) and k2j = −iωμ j σ j , for a time factor of e+iωt ). For a model
H1(1) have a singularity at the origin, such that their imaginary formed by N homogeneous layers, there will be N functions
parts tend to plus or minus infinity and their real parts are u j , j = 1, . . . , N.
discontinuous at that point, as shown in figure 2. The functions F (λ) and G(λ) will have two kinds of
The application of symmetry properties of the functions singularities: branch points and poles. They have exponentials
that form our integrands, as stated above, results in the with u j in their arguments, as well as expressions with u j in
following identities: quotients and products. Since Jn is analytical for every λ and
 ∞  Hn(1) has only one branch point at λ = 0 and is cut through
1 +∞
F (λ)J1 (λ r) dλ = F (λ)H1(1) (λ r) dλ, (7) the negative real axis, then for Im{λ} > 0 the only branch cuts
0 2 −∞
 ∞  will be those belonging to the square root in
1 +∞ 
G(λ)J0 (λ r) dλ = G(λ)H0(1) (λ r) dλ. (8) u j = λ2 + iωμ j σ j . (11)
0 2 −∞

3
J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

(a) (b)

(c) (d)
(1) (1)
Figure 3. H0 (λ) and H1 (λ), Im{λ} > 0. The thick red lines are J0 and J1 on the real line.

For each u j , the branch points are located on the straight the admittances of every layer beyond layer j and the infinite
line from the origin at 135o (if k2j = iωμ j σ j , the points are half-space at the base of the model:
on the line at 45o). The branch cutof the square root function Y j − Ŷ j+1
(u j ) starts at the point (−1 + i) ωμ j σ j /2 and follows the Rj = . (12)
Y j + Ŷ j+1
hyperbolic path Im{λ} = −ωμ j σ j /2Re{λ}, asymptotically
approaching √ the imaginary axis, as illustrated in figure 4 for The poles in this function are related to guided wave
the function λ2 + i. modes (Felsen and Marcuvitz 1994, Chew 1999) that are
Felsen and Marcuvitz (1994) show that the only cuts present when a resistive layer exists between two conductive
where the integrands will be multivalued functions are those ones, thus forming an open waveguide. The number of poles
depends on the frequency and on the thickness of the resistive
belonging to the two infinite regions at the top and bottom of
layer or layers in which there will be guided modes.
the layered models. In our models, the upmost medium is the
The companion paper (Howard and da Silva e Silva 2013)
infinite layer of air, with zero conductivity. Associated with
uses the guided mode condition on the fields and an asymptotic
this region is a branch cut along the imaginary axis starting at
method to show that the poles are located above the line formed
the origin of the system. Then, the only other multivalued cut by the branch points, as illustrated in figure 5. This is consistent
will be associated with the infinite region below the layers. with the estimated pole location given by Chew (1999).
The poles belong to the reflection coefficients (R) which Now the integrand function will vanish rapidly on the
are part of the functions F (λ) and G(λ). For the loop problem, integration path in the complex plane shown in figures 1
the reflection coefficient for the interface between layers j and 5. In the region between path (1) and the real axis the
and j + 1 is written in terms of the intrinsic admittance new integrand function has no singularities, as long  as path
(Y j = u j /iωμ j ) of one layer and of the surface admittance (1) passes under the branch points at (−1 + i) ωμ j σ j /2.
(Ŷ j+1 ) associated with the interface between the layers. The Cauchy’s integral theorem, then, applies and the integral over
latter is calculated as a recurrence formula that depends on that closed path, will be zero. Therefore, the integral over path

4
J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

(a) (b)
√ √
Figure 4. Real (a) and imaginary (b) parts of λ2 + i. The branch point is (−1 + i)/ 2.

Figure 5. Imaginary component of F (λ), for a three layer model, showing branch cuts and poles.

(1) in the direction shown in figure 1 is equal to that from On path (3) we have:
−∞ to −α over the real axis. The same can be said of the λ = λR (1 + i) − i α, (16)
integral over path (3): it will be the same as that from α to
 +∞
∞ on the real axis. So, after the application of the identities 7
and 8, the original improper integral becomes the sum of the I3 = (1 + i) F (λ)H1(1) (λ r) dλR . (17)
α
contributions of the three integration paths. Since the integrand Finally, on path (2) integration is on the real axis, so JN
vanishes quickly on paths (1) and (3), it is only necessary to can be used instead of HN(1) . Also, since the integrands are even
integrate on very short intervals in these paths. functions, only half the interval is needed:
The complex variable of integration λ has real and  α
imaginary components: I2 = 2 F (λ)J1 (λ r) dλ. (18)
0
λ = λ R + i λI . (13) The original integral is thus:
On path (1), in rectangular coordinates, parameterizing on  ∞
1
λR gives: F (λ)J1 (λ r) dλ = (I1 + I2 + I3 ), (19)
0 2
λ = λR (1 − i) − i α, (14) and likewise for all the equivalent expressions for the integrals
 in G(λ)J0 .
−α
I1 = (1 − i) F (λ)H1(1) (λ r) dλR . (15) To perform the actual integration, one must define the
−∞ three small paths following a few practical rules: α should have

5
J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

−5
|(E φ − E φa )/E φa | × 100
−5
10 10

−10 −6
10 10

Relative error (%)


E φ (V /m)

−15 −7
10 10

| Re{ E } | − Analytic
φ
−8
−20
10 | Re{ Eφ } | − Complex λ 10
| Im{ Eφ } | − Analytic
| Im{ Eφ } | − Complex λ
−25 −9
10 10
200 400 600 800 1000 1200 1400 1600 1800 200 400 600 800 1000 1200 1400 1600 1800 2000
r (m) r(m)
(a) (b)

−4 −2
|(H r − H ra )/H ra | × 100
10 10
| Re{ Hr } | − Analytic
| Re{ Hr } | − Complex λ
−4
−6 | Im{ H } | − Analytic 10
10 r

Relative error (%)


| Im{ Hr } | − Complex λ
−6
H r (A/m)

10
−8
10
−8
10

−10
10 −10
10

−12 −12
10 10
200 400 600 800 1000 1200 1400 1600 1800 200 400 600 800 1000 1200 1400 1600 1800 2000
r (m) r(m)

(c) (d)

0 −9
|(H z − H za )/H za | × 100
10 10
| Re{ H } | − Analytic
z
| Re{ Hz } | − Complex λ
−5
10 | Im{ H } | − Analytic
z
Relative error (%)

| Im{ H } | − Complex λ −10


z 10
−10
H z (A/m)

10

−15
10
−11
10
−20
10

−25 −12
10 10
200 400 600 800 1000 1200 1400 1600 1800 200 400 600 800 1000 1200 1400 1600 1800 2000
r (m) r(m)

(e) (f)

Figure 6. Relative differences between the field components of a DMV over a uniform half-space calculated with complex plane integration
and from the analytical expressions.

R the smallest possible value, to help the numerical integration


Tx on the central path, but it must not be too small, in order for
the integration on paths (1) and (3) to avoid the singularity
 at
ρ1 = 10 Ω.m 75 m the origin. Our practice has been to make α = ωμ j σm /2,
where σm is the highest conductivity in the model, therefore
ρ2 = 1000 Ω.m 50 m we start path (1) exactly under the furthest branch point in the
plane. The lengths of paths (1) and (3) must be enough for
the contribution of the remaining infinite length paths to be
ρ3 = 100 Ω.m
negligible.
On the paths (1) and (3) we apply the QWE algorithm to
Figure 7. Example model. Loop radius is R = 340 m. All results speed the convergence of a sequence of evaluations on small
were calculated at two frequencies, 0.1 Hz and 1000 Hz, at 1500 m sub-intervals. The calculation in each sub-interval is done with
from the centre of the loop. a 16 point Gauss–Legendre quadrature (Davis and Rabinowitz
1975), which is enough to ensure a relative error tolerance

6
J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

0.08 0.06

0.06 0.04

0.04

I m{integr and of E φ }
Re{integr and of E φ }

0.02

0.02
0
0
−0.02
−0.02

−0.04
−0.04

−0.06 −0.06

−0.08 −0.08
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Re{λ} Re{λ}
(a) (b)

0.3 0.3

0.2 0.2

I m{integr and of H r }
Re{integr and of H r }

0.1 0.1

0 0

−0.1 −0.1

−0.2 −0.2

−0.3 −0.3

−0.4 −0.4
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Re{λ} Re{λ}
(c) (d)
0.2 0.2

0.15 0.15

0.1 0.1
I m{integr and of H z }
Re{integr and of H z }

0.05 0.05

0 0

−0.05 −0.05

−0.1 −0.1

−0.15 −0.15

−0.2 −0.2
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Re{λ} Re{λ}
(e) (f)

Figure 8. Integrand functions on the real axis for the field components on the surface, at a distance of 1500 m and a frequency of 1000 Hz
for the model in figure 7.

of less than 10−9 and an absolute tolerance of 10−30 . In the 3. Validation


comparisons between the different methods, these are the same
demands imposed on the QWE method applied to the original In this section we show the results of applying the complex
integral on the real axis. On the path (2) the evaluation of plane integration with extrapolation for the fields of a VMD
equation (18) uses the same Gauss–Legendre quadrature with on the surface of a half-space of uniform resistivity of 1 
16 points.
m. In this case, there are analytical solutions for all three field
We have also successfully applied this method to calculate
components, as shown in Ward and Hohmann (1987). The plots
the fields from electric dipoles in the modelling of marine
CSEM data (Howard 2011), and to the integration of the loop’s in figure 6 show the amplitude of the fields for this model and
magnetic field on the area of a receiver loop, in which case the the relative differences between those calculated by numerical
integrand has extremely slow convergence due to the product integration and those given by the analytical solutions. These
of three Bessel functions (de Carvalho et al 2011). results were generated for a frequency of 50 Hz, and the fields

7
J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

0.08 0.08
path 2 path 2
0.06 0.06

I m{integrand of E φ }
Re{integrand of E φ }

0.04 0.04

0.02 path 1 0.02 path 1 path 3


path 3
0 0

−0.02 −0.02

−0.04 −0.04

−0.06 −0.06

−0.08 −0.08
−0.05 −0.04 −0.03 −0.02 −0.01 0 0.01 0.02 0.03 0.04 0.05 −0.05 −0.04 −0.03 −0.02 −0.01 0 0.01 0.02 0.03 0.04 0.05
Re{λ} Re{λ}
(a) (b)

0.3 0.3

0.2 0.2

I m{integrand of H r }
Re{integrand of H r }

0.1 path 1 path 3 0.1


path 1 path 3

0 0

−0.1 −0.1

−0.2 −0.2
path 2
−0.3 path 2 −0.3

−0.4 −0.4
−0.05 −0.04 −0.03 −0.02 −0.01 0 0.01 0.02 0.03 0.04 0.05 −0.05 −0.04 −0.03 −0.02 −0.01 0 0.01 0.02 0.03 0.04 0.05
Re{λ} Re{λ}
(c) (d)
0.3 0.3

0.2 0.2
I m{integrand of H z }
Re{integrand of H z }

0.1 0.1
path 1 path 3 path 1 path 3
0 0

−0.1 −0.1

−0.2 −0.2
path 2 path 2
−0.3 −0.3

−0.4 −0.4
−0.05 −0.04 −0.03 −0.02 −0.01 0 0.01 0.02 0.03 0.04 0.05 −0.05 −0.04 −0.03 −0.02 −0.01 0 0.01 0.02 0.03 0.04 0.05
Re{λ} Re{λ}
(e) (f )

Figure 9. Integrand functions for the field components on the surface, at a distance of 1500 m and a frequency of 1000 Hz for the model in
figure 7.

calculated at distances from 10 up to 2000 m from the source. Hohmann (1987), requires the numerical evaluation of the
The agreement is very good for all three components. inverse Hankel transform, in the form shown in equations (3)
and (4):
 ∞
4. Application Eφ (r, z, ω) = Êφ (λ, z, ω)λJ1 (λ a)J1 (λ r) dλ, (20)
0
Consider the calculation of the fields generated by a horizontal  ∞
∂ Êφ (λ, z, ω)
circular loop of current placed on the surface of a stratified Hr (r, z, ω) = λJ1 (λ a)J1 (λ r) dλ, (21)
0 ∂z
earth. Although geophysical data are usually measured on the  ∞
surface, accurate calculations of the electromagnetic field in Hz (r, z, ω) = − Êφ (λ, z, ω)λ2 J1 (λ a)J0 (λ r) dλ, (22)
arbitrary points inside the 1D layered earth are important, 0
because these are used as primary fields in numerical methods where a is the loop radius, J0 and J1 are Bessel functions of
in the modelling of 2.5D and 3D structures. the first kind, and Êφ (λ, z) is a frequency dependent function
In cylindrical coordinates, the horizontal loop generates that carries the information about the source and the model in
the components Eφ , Hr and Hz . The calculation of those the transform domain, including all reflections and refractions
components, as described in Ryu et al (1970) and Ward and between layers (Ward and Hohmann 1987). One important

8
J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

−11 −8
x 10 x 10
−7.85 −1
Guptsarma (140 abscissae) Guptsarma (140 abscissae)
Anderson (801 abscissae) Anderson (801 abscissae)
Complex λ Complex λ
−7.9 Real λ −1.01 Real λ

I m{E φ } (V /m)
Re{E φ } (V /m)

−7.95 −1.02

−8 −1.03

−8.05 −1.04
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
z (m) z (m)

(a) (b)
−6 −8
x 10 x 10
4 7
Guptsarma (140 abscissae) Guptsarma (140 abscissae)
Anderson (801 abscissae) 6 Anderson (801 abscissae)
Complex λ Complex λ
3
Real λ 5 Real λ

I m{H r } (A/m)
Re{H r } (A/m)

2 4

3
1 2

1
0
0

−1 −1
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
z (m) z (m)

(c) (d)
−6 −8
x 10 x 10
−7.8 −3.5
Guptsarma (120 abscissae)
−8 Anderson (801 abscissae)
Complex λ
−3.6
−8.2 Real λ
I m{H z } (A/m)
Re{H z } (A/m)

−8.4 −3.7
−8.6

−8.8 −3.8

−9 Guptsarma (120 abscissae)


−3.9 Anderson (801 abscissae)
−9.2 Complex λ
Real λ
−9.4 −4
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
z (m) z (m)

(e) (f)

Figure 10. Calculated field components for the model in figure 7 at 0.1 Hz. The legends real λ and complex λ refer to the integration with
the QWE algorithm applied on the real axis and on the paths in the complex plane respectively.

point about Êφ (λ, z) is that it is an even function of λ. It is, in the slowly converging curves seen on the real axis, now we
fact, a function of λ2 . have very fast convergence to zero on the two complex plane
In this example, the integration procedure is applied to paths. In this example, after an interval of only 0.038 m−1
calculate the fields in the model shown in figure 7. The on paths (1) and (3), the integration converged within the
radius of the transmitter loop is 340 m. For thismodel, at established criteria.
a frequency of 1000 Hz, the values assumed by ωμ j σ j /2 Note that the functions are neither odd nor even. They are
are, approximately, 2.0 × 10−2 , 2.0 × 10−3 and 6.3 × 10−3 . shown after the substitution of H0(1) and H1(1) for J0 and J1 , so
Therefore, following our rule of thumb, α = 2.0 × 10−2 . At both the real and the imaginary parts are the sum of odd and
0.1 Hz, the values are two orders of magnitude smaller, with
even functions, hence the lack of symmetry.
α = 2.0 × 10−4 .
Figures 10 and 11 show the results for this model.
Figure 8 shows the behaviour of the integrand functions
in the calculation of the Eφ , Hr and Hz components, on the real The fields calculated by complex plane integration with
axis, with z = 0, r = 1500 m and a frequency of 1000 Hz. extrapolation are compared with those calculated by real axis
The modified integrands for these same components, on the integration using three different methods. The first one is the
three paths in the complex plane, are shown in figure 9. Here QWE algorithm, as presented by Key (2012); the other two
the motivation to go into the complex plane is clear: instead of methods are the filters presented by Anderson (1979) and by

9
J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

−7 −7
x 10 x 10
0.5 2.5

0 2

−0.5
1.5

I m{E φ } (V /m)
Re{E φ } (V /m)

−1
1
−1.5
0.5
−2
0
−2.5 Guptsarma (140 abscissae) Guptsarma (140 abscissae)
Anderson (801 abscissae) Anderson (801 abscissae)
−3 −0.5
Complex λ Complex λ
Real λ Real λ
−3.5 −1
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
z (m) z (m)
(a) (b)
−7 −6
x 10 x 10
20 0
Guptsarma (140 abscissae)
Anderson (801 abscissae)
Complex λ
15 −0.2
Real λ

I m{H r } (A/m)
Re{H r } (A/m)

10 −0.4

5 −0.6

Guptsarma (140 abscissae)


0 −0.8
Anderson (801 abscissae)
Complex λ
Real λ
−5 −1
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
z (m) z (m)
(c) (d)
−7 −8
x 10 x 10
14 20
Guptsarma (120 abscissae) Guptsarma (120 abscissae)
12 Anderson (801 abscissae) Anderson (801 abscissae)
Complex λ Complex λ
15
10 Real λ Real λ
I m{H z } (A/m)
Re{H z } (A/m)

8 10

5
4

2
0
0

−2 −5
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
z (m) z (m)
(e) (f)

Figure 11. Calculated field components for the model in figure 7 at 1000 Hz. The legends real λ and complex λ refer to the integration with
the QWE algorithm applied on the real axis and on the paths in the complex plane respectively.

Guptasarma and Singh (1997). The limitations of the filters 140 points for the integral with J1 and in 120 points for the
are evident in these figures. integral with J0 . In figure 12, we compare the number of
In all cases, the extrapolation method on the real axis and integrand evaluations required by the QWE method to achieve
on the paths in the complex plane were applied with the same the same precision in both applications, with a convergence
requirement of convergence criteria, and they both generated criterion of relative tolerance from one iteration to the next of
the correct responses with the same level of precision. In order less than 10−9 , and an absolute tolerance of less than 10−30 , to
to achieve the same results, however, the integrations with the calculate the results shown in figure 11 at the frequencies of
QWE method applied on the real axis and on the complex 0.1 Hz and 1000 Hz.
plane do not require the same computer time. Since the integrands converge to zero much more rapidly
In order to measure the performance of these two on the complex plane paths than on the real axis, the
algorithms, we count the number of evaluations of the QWE process requires fewer evaluations, which implies less
integrand functions in each case. As a reference in this computer time. Table 1 shows the total time required for the
comparison, note that the Anderson filter evaluates the two QWE integration methods to compute all the results of
integrands in 801 points for each integral in J0 and J1 and figures 10 through 12, using FORTRAN routines, running on
the Guptasarma and Singh filter evaluates the integrands in Mac OS X notebook with a 2 GHz processor.

10
J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

16 point sub-intervals. Frequency: 0.1 Hz

352
336

I ntegrand evals
320 E φ (real λ)

288

256
H z (real λ)
224 H r (real λ)
E φ , H r , H z (complex λ)
176

0 20 40 60 80 100 120 140 160 180 200


z (m)
(a)
16 point sub-intervals. Frequency: 1000 Hz

H z (real λ)
416
400 H r (real λ)
384
I ntegrand evals

368
352
E φ (real λ)

E φ , H r , H z (complex λ)
208
0 20 40 60 80 100 120 140 160 180 200
z (m)
(b)

Figure 12. Number of evaluations of the integrand functions in the two integration methods using QWE: on the real axis and in the complex
plane.

Table 1. Time in seconds for both QWE integration methods to filters, even when the latter are accurate. So, if computer time
generate the data shown in figures 10, 11 and 12. is a concern, one must choose carefully the right tool for each
0.1 Hz 1000 Hz case.
Method Eφ Hr Hz Eφ Hr Hz
5. Conclusion
QWE (16 pts) 0.1306 0.1485 0.1318 0.1905 0.1982 0.1978
QWECP (16 pts) 0.1007 0.1034 0.1028 0.1035 0.1035 0.1032
The method discussed here is a viable alternative for evaluation
of integrals involving Bessel functions, as in the applications of
Note that computer time is always small for any layered Hankel transforms. It is rigorous and valid generally. The only
medium problem like this, so that it is not a concern in itself, approximations are in the numerical method used to evaluate
except perhaps when these calculations need to be repeated the integrals on the three paths.
a very large number of times, such as when they are used as The modelling of the electromagnetic multi-frequency
primary fields in 3D problems. method requires the fields from large loops to be calculated in
A final remark about the digital filters: they have been a large dynamic range of observation points and frequencies.
used successfully for a long time in many different problems A dipole approximation is not possible, because the problem
in geophysical applications. In most cases they are efficient requires fields close to the source, as well as at distances up
and easy to use. The large loop problem studied here presents to 10 km from it. In these situations the method performed
situations which are not well approximated by the filters, but well, and in the most extreme cases the use of quadrature with
it must be recognised that these are very particular cases. On extrapolation was the only method that achieved satisfactory
the other hand, it has been shown in Key (2012) that in many results, being most efficient when applied to the integration on
situations the QWE algorithm can be more efficient than the the complex plane.

11
J. Geophys. Eng. 11 (2014) 015004 V da Silva e Silva et al

Acknowledgments Dias C A, Sato H K and de Lima O A L 2005 Multi-frequency EM


method for hydrocarbon detection and for monitoring fluid
The authors thank the National Institute of Science and invasion during enhanced oil recovery 2005 SEG Annu.
Meeting (Houston, TX, 6–11 Nov.) pp 602–4 (expanded
Technology of Petroleum Geophysics—INCT/GP and abstracts)
projects 0050.0056738.10.9-UFBA/FAPEX/PETROBRAS Felsen L B and Marcuvitz N 1994 Radiation and scattering of waves
and 0050.0046724.08.9-UFPA/FADESP/PETROBRAS, both IEEE Press Series on Electromagnetic Waves (Piscataway, NJ:
part of the Network of Applied Geophysics at PETROBRAS, IEEE)
for supporting this research. Fullagar P K and Oldenburg D W 1984 Inversion of horizontal
loop electromagnetic frequency soundings Geophysics
49 150–64
References Guptasarma D and Singh B 1997 New digital linear filters for
Hankel J0 and J1 transforms Geophys. Prospect.
Abramowitz M and Stegun I A 1965 Handbook of Mathematical 45 745–62
Functions (New York: Dover) Howard A Q Jr 2000 Petrophysics of magnetic dipole fields
Anderson B, Safinya K A and Habashy T 1986 Effects of in an anisotropic earth IEEE Trans. Antennas Propag.
dipping beds on the response of induction tools 61st Annu. 48 1376–83
Technical Conf. and Exhibition of the Society of Petroleum Howard A Q Jr 2011 On numerical integration of sea bed logging
Engineers 3 29–36 em models CISBGf: 12th Int. Congr. of the Brazilian
Anderson W L 1979 Numerical integration of related Hankel Geophysical Society (expanded abstracts)
transforms of orders 0 and 1 by adaptive digital filtering Howard A Q Jr and da Silva e Silva V 2013 Complex plane
Geophysics 44 1287–305 integration in the modeling of electromagnetic fields in layered
Chave A D 1983 Numerical integration of related media: part 2. Pole locations J. Geophys. Eng. 11 015005
Hankel-transforms by quadrature and continued fraction Key K 2012 Is the fast Hankel transform faster than quadrature?
expansion Geophysics 48 1671–86 Geophysics 77 F21–F30
Chew W C 1999 Waves and Fields in Inhomogeneous Media (IEEE Machado M V B 2009 Estudo analı́tico e aplicação do método
Press Series on Electromagnetic Waves) (Piscataway, NJ: eletromagnético a multi-frequência ao mapeamento e
IEEE) identificação de fluidos e reservatórios continentais de petróleo
Christensen N B 1990 Optimized fast Hankel transform filters PhD Thesis Universidade Estadual Norte Fluminense Darcy
Geophys. Prospect. 38 545–68 Ribeiro (in Portuguese)
Davis P J and Rabinowitz P 1975 Methods of numerical integration Poddar M 1983 A rectangular loop source of current on
Computer Science and Applied Mathematics (New York: multilayered earth Geophysics 48 107–9
Academic) Ryu J, Morrison H F and Ward S H 1970 Electromagnetic fields
de Carvalho P R, Régis C R T and Howard A Q Jr 2011 Magnetic about a loop source of current Geophysics 35 862–96
point dipole versus finite-size circular loop models in coaxial Singh N P and Mogi T 2005 Electromagnetic response of a large
and coplanar two coil arrays in induction tools CISBGf: 12th circular loop source on a layered earth: a new computation
Int. Congr. of the Brazilian Geophysical Society (expanded method Pure Appl. Geophys. 162 181–200
abstracts SBGf) Ward S H and Hohmann G W 1987 Electromagnetic theory for
Dias C A, de Lima O A L, Sato H K and Moraes J A C 2006 geophysical applications Electromagnetic Methods in Applied
Contribution to oil exploration and development—a successful Geophysics (Investigations in Geophysics vol 1) (Tulsa, OK:
inductive multi-frequency EM survey on-shore Brazil EAGE Society of Exploration Geophysicists) chapter 4
68th Conf. and Exhibition (Vienna, Austria, 12–15 June) Watson G N 1966 A Treatise on the Theory of Bessel Functions
pp 61–5 (Cambridge: Cambridge University Press)

12

Das könnte Ihnen auch gefallen