108

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

Journal of Zhejiang University-SCIENCE A (Applied Physics & Engineering)

ISSN 1673-565X (Print); ISSN 1862-1775 (Online)
www.zju.edu.cn/jzus; www.springerlink.com
E-mail: [email protected]

A fast integration method for translating-pulsating source

Greens function in Bessho form*
Chao-bang YAO, Wen-cai DONG
(Department of Naval Architecture Engineering, Naval University of Engineering, Wuhan 430033, China)
E-mail: [email protected]; [email protected]
Received June 16, 2013; Revision accepted Oct. 7, 2013; Crosschecked Jan. 16, 2014

Abstract: The singularities and oscillatory performance of translating-pulsating source Greens function in Bessho form were
analyzed. Relative numerical integration methods such as Gaussian quadrature rule, variable substitution method (VSM), and
steepest descent integration method (SDIM) were used to evaluate this type of Greens function. For SDIM, the complex domain
was restricted only on the -plane. Meanwhile, the integral along the real axis was computed by use of the VSM to avoid the
complication of a numerical search of the steepest descent line. Furthermore, the steepest descent line was represented by the
B-spline function. Based on this representation, a new self-compatible integration method corresponding to parametric t was
established. The numerical method was validated through comparison with other existing results, and was shown to be efficient
and reliable in the calculation of the velocity potentials for the 3D seakeeping and hydrodynamic performance of floating structures moving in waves.
Key words: Translating-pulsating source Greens function, Oscillatory performance, False singularities point, Steepest descent
integration method (SDIM), Variable substitution method (VSM)
doi:10.1631/jzus.A1300209
Document code: A
CLC number: U661.32

1 Introduction
The frequency-domain Greens function with
forward speed in 3D deep water was derived by
Haskind (1953), and its physical meaning is the velocity potential of the field point produced by the unit
translating and pulsating source. It has been widely
used in computing wave-induced ship motions and
loads (Bailey et al., 2001; Chen and Wu, 2001; Fang
and Chan, 2002; Du et al., 2005; 2012; Zakaria, 2006;
2009; Zong and Qian, 2012). The advantages of this
Greens function are an automatic fulfillment of the
linearized free surface and radiation conditions, and
the absence of mesh on the free surface avoiding
*

Project supported by the National Natural Science Foundation of

China (No. 50879090), and the Key Research Program of Hydrodynamics of China (No. 9140A14030712JB11044)
Zhejiang University and Springer-Verlag Berlin Heidelberg 2014

reflection on the boundary. However, it is difficult to

ensure calculation efficiency and accuracy. The numerical difficulties of the problem can essentially be
traced to the complexity of the mathematical expression for the Greens function, or more precisely of any
of the three available alternative expressions for this
fundamental function. Xu and Dong (2011) discussed
three other forms of this function such as Bessho form
(Bessho, 1977), Michell form (Miao et al., 1995) and
Havelock form (Rahman, 1990), which were obtained
by using elementary mathematics. The Havelock
form is a single integral. There are some difficulties in
calculation such as infinite discontinuity when the
denominator of the integrand equals zero, a jump
discontinuous node in the exponential integral function of the local disturbance part, and high-frequency
oscillatory performance of the far-field wave-like
part. The Michell form is not suitable for numerical
analysis because part of the function is a double

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

integral. The Bessho form is considered as the most

suitable for calculations because it is expressed by a
single integral of the elementary functions in the
complex domain. This single integral expression has a
definite advantage over the others, which are expressed by using the exponential integral functions E1
from the point of view of computational accuracy and
speed (Iwashita and Ohkusu, 1989; 1992; Du and Wu,
1998; Du et al., 2005). On the other hand, the numerical integration must be performed in the complicated complex domain, and there are still some
difficulties in calculation as follows: (1) an infinite
discontinuity exists when the denominator of the
integrand is zero; (2) high frequency oscillatory performance of the integrand is complicated.
Some numerical integration methods have been
studied to dispose of such singularities and high frequency oscillatory performance. Iwashita and Ohkusu
(1989; 1992) developed a special algorithm, named
numerical steepest descent method, for evaluating
Bessho form translating-pulsating source Greens
function. Originally this method was obtained by
transforming the complex -plane to the complex
M-plane, where the integration range was mapped to
the half-infinite region. Following this transformation, Du (1996) and Du and Wu (1998) established
a fast algorithm for calculating this form of Greens
function by employing the steepest descent path
method together with a self-adaptive integration
method. Brument and Delhommeau (1997), Maury
(2001), and Maury et al. (2003) further studied the
choice of step, start points, and common use of many
descent lines for the steepest descent method. The
steepest descent integration method (SDIM) on the
M-plane makes the numerical search of the steepest
descent line easy when approaches /2. However,
the presence of new branch cuts must be taken into
consideration and this sometimes complicates the
numerical search of the steepest descent line. To improve this computational method, Iwashita (1997)
directly performed the integration on the -plane
without any transformation to other complex planes.
However, to do the calculation, the information about
the saddle points and the cross points between the real
axis and the steepest ascent line must be analyzed to
avoid many wrong paths when the start point is located on the real axis. This may be sometimes a waste
of time.

109

In this paper, the singularities, oscillatory performance and their contribution factors of the Bessho
form Greens function were analyzed systematically,
aiming to propose a fast numerical integration method
based on the SDIM. The efficiency and accuracy of
these integration methods were validated by numerical
results.

2 Single integral expression of translatingpulsating source Greens function in Bessho

form
Assuming the translating-pulsating source is
traveling at a forward speed U and oscillating at a
frequency e, (, , ) is the source point and (x, y, z)
is the field point, then the 3D translating-pulsating
source Greens function can be expressed as
G ( x, y, z; [ ,K , ] )

1 1 1

 G ,
4S r r1

G ( x, y, z; [ ,K , ] )

i
K 0T ( X , Y , Z ),
2S

(1a)
(1b)

where
T ( X ,Y , Z )

dT

T2

T2

H (T )dT =

1  4W cos T
T2
dT

[k2 ek2Z  k1ek1Z sgn c]

1  4W cos T
 k1Mk1 ( X , Y , Z , k1 )sgn c],
1

T1

r1

( x  [ )2  ( y  K )2  ( z # ] )2 ,

k1

k2

Z
U

[k2Mk2 ( X , Y , Z , k2 ) (2a)

(2b)

1
(1  2W cos T r 1  4W cos T ),
2 cos 2 T

Z  i( X cos T  Y sin T )
X2 Y2, J

T2

S
 M  iH , H
2

Z  iU cos(T  J ), (2d)

arg( X  iY ),

1
arc cos , 4W ! 1,

4W

iar cosh 1 , 4W  1,
4W

(2c)

Z
ar sinh
X2 Y2

(2e)

(2f)

(2g)

110

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

X
(2h)
arc cos
,
2
2
X Y
Mk j ( X , Y , Z , k j ) exp[k j Z  ik j ( X cos T  Y sin T )], (2i)
U Ze
,M
g

K0

g
,X
U2

K 0 ( x  [ ),

K0 y  K , Z

T1

S  D ,

(2j)

K 0 ( z  ] ),

The integration methods for T presented in the

following are also suitable for its three partial derivatives as given in Eq. (4), because no other calculation
difficulties are added.
(a) S  D

S / 2

S / 2 + M

S / 2

S / 2 + M

Re(T )

(2k)
(2l)

S
S

1,  2 d Re(T ) d 2 ,
(2m)
sgn c sgn[cos(Re(T ))]
1,  S d Re(T )   S ,

where r and r1 are distances to source point and its

symmetry with respect to the free surface, respectively; k1 and k2 are non-dimensional wave numbers
of the propagation waves; is Brard parameter; K0 is
wave number; and g is the gravitational acceleration.
Integral paths of Bessho form Greens function
are shown in Fig. 1, where the symbols are defined as
shown in Eq. (2).
Despite the integral limits depending on the coordinates of the field points, differentiation of T with
respect to x, y, and z is not as complicated as anticipated. The derivatives are listed as follows:

1

Gx

sgn( y  K ) G

G
1
z

sgn( y  K ) T ( X , Y , Z )

i cos T
1
i sin T
(3)

1  4W cos T
1

 k12 e k1w sgn c]dT  T0 ( X , Y , Z ),

S
  M  iH
2
D S

K0

[k22 e k2 w

Y X 2  Y 2  Z 2  iX Z


T0 ( X , Y , Z ) 
Y2  Z2

Y

X Z
i
2

2
( X 2  Y 2 ) X 2  Y 2  Z 2 (4)
X Y

Y Z
X

.
2
i
X Y2
(X 2  Y 2) X 2  Y 2  Z 2

2
2
2
X Y  Z

S / 2 + M  iH

(b)

S

S  D

Re(T )

S / 2 + M  iH

Fig. 1 Integral path: (a) corresponds to >0.25; (b) corresponds to <0.25

3 Characteristics of the integrand of T

3.1 Characteristics of ki
ki is the non-dimensional wave number of the
propagation wave generated by the translatingpulsating source whose independent variables are
and . ki is an important parameter of the integrands
which influence performances of the propagation
parts. Meanwhile, ki is also dependent on the definition domain of , which is defined by the relative
position of field and source points. Typical function
images of ki are presented in Fig. 2: T
[S, S / 2] [S / 2, 0] if W d 0.25 and T
[T1 , S / 2] [S / 2, 0] if W ! 0.25 .

Some findings for the characteristics of ki can be

drawn as follows: (1) for T S / 2, k1 / W 2 presents
the tendency to infinity; as a result, the integrand
oscillates with high frequency as  approaches
S / 2, which may be a big obstacle for numerical
integration. (2) if T o S / 2, then k2 / W 2 o 1,
terms in k2 do not oscillate with high frequency, and
numerical integration is straightforward. Due to their
behavior, terms in k1 and k2 are computed differently.
From the point of numerical calculation, under
conditions of T o S / 2, k2 o W 2 , and k2 can be

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

calculated according to Eq. (5), if T r S / 2  N ( N

where

is infinitely small),

W (1  2a  5a  14a  42a ),
a W cos T .
2

The integral limits of Eq. (2a) depend on and ,

the lower limit 1 is complex or a real number corresponding to <0.25 or >0.25. When X0, [0,
/2]; when X<0, [/2, ].
To investigate the behavior of the integrands, we
separate the integral over  into different parts according to the sign of terms sgnc.
2

T2

j 1

1  4W cos T

(5)

3.2 Singularity and oscillatory behavior of the

integrand

T ( X ,Y , Z )

k jM k j c j

H j (T )

H j (T )dT

2 S / 2 S / 2 T2
[ T1  S  S / 2 ]H j (T )dT , W  0.25, (6)
j 1
2
[ S / 2  T2 ]H (T )dT , W ! 0.25,
T1 S/ 2 j

j 1

-2
Im(T1)
(b)

0.0

-0.8
-1.6

100

-2.4
0

-3

-2

-1

-S

-S

(c)

4
(b)

W<0.25
W=0.25
W>0.25

1
Re(H1)

3
k2/W

S

0.8
Re(H1)

k1/W

200

1.6

W<0.25
W=0.25
W>0.25

300

1, j 2,

 sgn c, j 1.

(a)

2.4
(a)

and c j

Fig. 3 and Fig. 4 show images of Hj() when

T [T1 ,T 2 ], for =0.2, X=500, Y=1, and Z=0.2. The
performance of the integrand in this case can be
summarized as follows.
(1) It is certain that 1 is an infinite discontinuity
and values of terms in k1 and k2 change violently as 
approaches 1.

Re(H1

k2

111

0
-1

1
0

-2
S/2

-3

-2

-1

Fig. 2 Typical images of k1/2 (a) and k2/2 (b)

Re(T2)

Fig. 3 Typical images of integrands corresponding to k1

(=0.2)
(a) [Im(1), ]; (b) [, /2]; (c) [/2, Re(2)]

112

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

k Z

other part e j is an attenuation term (kjZ<0) and

determines amplitudes of Hj. When | k j Z | is getting

(a)

larger, the attenuation speed of Hj becomes higher. As

for H2, the value of k2 is not as large as shown in
Fig. 2, so only when the parameter is very large can
H2 act as high-frequency oscillatory. In addition, attenuation performance of H2 is not obvious because
values of | k2 Z | are not large either. As far as H1 is

Re(H2)

-1

-2
Im(T)

S

concerned, k1 is very large when T o S / 2, so H1

becomes high-frequency oscillatory not only when
is very large but also when T o S / 2. However, H1
attenuates very fast when  approaches S / 2 since
| k1 Z | is also very large and H1 becomes zero when

0.2
(b)

Re(H2)

0.1
0.0

S / 2.

-0.1

4 Numerical integration methods

-0.2

-S

-S2

0.04

4.1 Integration methods for the singularities at

terminals

(c)

From the analysis above, we can see that 1 is an

infinite jump discontinuous node of terms both in k1
and k2. Gaussian quadrature rule (Abramowitz and
Stegun, 1964) is used in this study to eliminate such a
node. However, this rule is not suitable to calculate
the oscillatory function, so we apply it only within the
domain [1, d], where d is defined by

Re(H2)

0.02
0.00
-0.02
-0.04
S/2

Re(T2)

Fig. 4 Typical images of integrands corresponding to k2

(=0.2)
(a) [Im(1), ]; (b) [, /2]; (c) [/2, Re(2)]

(2)

When T [S, S / 2] [S / 2,T 2 ],

the

performance of Mk j is oscillatory because of the

complex exponential function e

k jZ

(j=1, 2), which

Im(k jZ

In practice, 9

T T1

frequency and, the larger it is, the higher the frequency is. It can also be concluded that the value of
k j U determines whether Mk j is higher-frequency
oscillatory with the condition of cos(T  J ) d 1. The

T Td

)  9.

(7)

S / 10 is proved to be adequate.

By performing m T  S, t cos m and applying the Gaussian quadrature rule to Hj() when its
interval is [1, d], the integral can be obtained approximately as follows:

k j Z  ik j U cos(T J )

can be written as e
. The term
k j U cos(T  J ) determines the value of the oscillatory

)  Im(k j Z

Td

H (T )dT

where b

Zn
tn

P (t )

(t  a)(b  t )

dt | P (tn )'Zn , (8)

n 1

cos(T1  S) 1 / (4W ),

(2n  1)S / (2m),

Nn

cos Zn ,

cos(T d  S),

'Zn

S / m,

(a  b) / 2  (b  a )N n / 2, and m is the discrete

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

step of the interval. (t) can be written as

k jZ

P (t )

ik j e
ta
, W  0.25,

2
4W (t  1)

k jZ
ta
k je
, W ! 0.25.

2
4W (t  1)

(9)

The integral may be obtained by using the variable substitution method which will be discussed in
detail as below, when the interval is [T d , S] (<0.25)
or [T d , S / 2] (>0.25).
4.2 Integration methods for within the domain
of [1, /2]
It is inefficient and sometimes even impossible
to calculate integrals in k1 and k2 by using usual integration methods because oscillation characteristics
of these integrals are extremely complicated. The
steepest descent method is often used to compute
terms in k1; however, when the steepest descent line
starts from the real axis, the search algorithm of these
lines are often difficult and complicated. In this study,
a method based on a variable substitution is introduced for the calculation.
4.2.1 Integration method for on the real axis
If the interval for Hj() is the domain of [e, s],
the integral of Hj() can be written as Eq. (10) by
performing Aj k jZ.
Ts

H j (T )dT

A j (Ts )
A j (T e )

kj

Aj

1  4W cos T Acj

e dAj , (10)

Ts

where
dk j W sin T
(1) j 1
1


dT
cos 2 T
1  4W cos T
if T z rS / 2,

 2k j tan T , (12a)

dk1
f, if T rS / 2,
dT
dk2
r2W 3 , if T rS / 2,
dT
dk i
dZ
iU sin(T  J ),
l
Z,
dT
dT
dk j

d U
cos(T  J )  k j sin(T  J ) .
T
d

(12b)
(12c)
(12d)
(12e)

Integrands of Eq. (11) can be written as f j e j ,

where fj is a complex function and can be expressed as
follows:
fj

k j (l  id )
1  4W cos T (l 2  d 2 )

(13)

Assuming that the discrete sequences of [e, s]

are T1 , T 2 , ..., T m , the relative sequences of Aj are
Aj(1), Aj(2), , Aj(m).
If a linear function about Aj is used to approximate fj between Aj(n) and Aj(n+1) ( 1 d n d m  1 ), the
integral In between the two points can be expressed as
Eq. (14) through integration by parts
In

[ f j (T n1 )e


dk i
dZ
Z
kj.
dT dT
dT
dA j
Performing
l  id , where l is a real part
dT
and d is an imaginary part. Substitute it into Eq. (10)
and multiply it by lid, then Eq. (10) can be converted
into the following expression:

where

113

A j (T n1 )

 f j (T n )e

[ f j (T n1 )  f j (T n )][e

A j (T n )

A j (T n1 )

Aj (T n1 )  Aj (T n )

e

]
A j (T n )

(14)

dA j

H j (T )dT

A j (Ts )
A j (T e )

k j (l  id )

1  4W cos T (l  d )
2

e j dAj ,

(11)

Thus, the sum of In approximates to the integral

of Hj() ([e, s]). The accuracy of this integration
method is determined by the performance of fj and its
relative interval discrete method. It is not difficult to
prove that fj is continuous within the integration
interval.
The completed variable substitution has transformed the oscillatory function into a formula easy for
integration. It should be pointed out that values of fj
vary violently in a narrow zone near relevant f when
d equals zero. f is defined to be the root of equation

114

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

d=0, which can be simplified as Eq. (15) by some

mathematical derivations. It can be easily obtained
that the root number of Eq. (15) is smaller than 2.

1
tan T 1  (1) j
1  4W cos T

 tan(T  J )

0. (15)

Fig. 5a gives typical curves of Re[fj] and Im[fj]

when no root is existing in Eq. (15). The curves are
smooth and easy to discrete. Typical curves of Re[fj]
and Im[fj] with one root existing in Eq. (15) are given
in Fig. 5b and the characteristics of these curves are:
(1) The curve shape of Im[fj] in the narrow zone
near f is leptokurtic
(2) Values of Re[fj] change violently in the narrow zone around f, and f looks like a jump discontinuous node but it is a false appearance here because
fj is continuous at the point of f. This false appearance is defined as a false bizarre phenomenon and the
respective f is defined as a false singularity. The false
bizarre phenomenon should be taken into account to
improve the integration efficiency and accuracy during discrete treatment of the integral interval. At this
point, the local refinement of the integral steps technique should be applied.
0.4
0.3

4.2.2 Integration method for [1, ] when <0.25

The Gaussian quadrature rule is applied to
evaluate the integration within the domain [1, d] (d
is defined as above), when <0.25. However, within
the domain of [d, ] the integrands of Hj() also
have complicated oscillatory characteristics when
has a large value or the locations of source points and
field points change. At this point, the method based
on variable substitution is used again.
If the interval is [e2, s2], then the integral of
Hj() can be calculated according to Eq. (16) by
performing Aj=kj.
Ts 2

Ts 2

e2

dk j

Real part

dT

hij dT

dA
dT

fj

-0.1
-0.2

(a)

Ts

Tf

20
Imaginary part

10

Real part

1  4W cos T Acj

e j dAj .

(16)

k j (l  id )

A j (Ts 2 )
A j (T e 2 )

1  4W cos T (l 2  d 2 )

e j dAj , (17)

(1) j 1
W sinh m
1 
 2k j tanh m,
2
cosh m
1  4W cosh m
dk j

Im(T  S), if T z rS / 2,

Z  kj

dZ
dT

dk j

( Z  Y sinh m)
dT
dk j

 k jY cosh m  iX
cosh m  k j sinh m ,
dT

dk j
l
( Z  Y sinh m)  k j Y cosh m,
dT
dk j

d iX
cosh m  k j sinh m .
dT

dT

fj

The integration method is the same as that discussed in Section 4.2.1.

-10

4.3 Integration methods for within the domain

of [/2, 2]

-20
(b)

-30

0.0

A j (T e 2 )

wherein

Imaginary part

0.1

kj

A j (Ts 2 )

Similar to the method in Section 4.2.1, the

foregoing Eq. (16) can be converted into the following expression:

0.2

-0.3

e2

hij dT

Ts

Tf

Fig. 5 Typical characteristics of fj: (a) no root existing;

(b) one root existing

For this numerical simulation within the domain

[/2, 2], we divide the complex -plane into regions

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

as shown in Fig. 6, which is proposed by Iwashita

(1997). We consider a different treatment of the integration path for each case, and choose different integration methods. In Fig. 6, is a relatively small value
and set as =/16 in actual calculation.
(a) For the condition of 10 including the case
of X=Y=0
The integrand tends to zero as tends to infinite.
We can therefore neglect the integration from 2 to
10i as the order of the integrand is less than O(108)
in the magnitude at 2=10i. The integration path is
taken as straight lines shown in Fig. 6a, because the
steepest descent line from, i.e., =10i is regarded as
the imaginary axis and the real axis towards =/2.
At this point, we calculate the integration along the
real axis by the VSM, which is different from Iwashita (1997)s method and proved to be efficient.
(b) For the condition of 4<10
The path from 2 is taken to be close to the imaginary axis by using the B-spline discussed by Yao
and Dong (2013). The cross points between the dotted
line and the imaginary axis are used for the control
points of the B-spline. Here, the integration along the
real axis is also calculated by the VSM to avoid the
complication of numerical search of the steepest descent line when the start point is located on the real
axis.
(c) For the condition of <4
If Re(2)/16, the path is toward =/2;
otherwise, if Re(2)/16, the path is toward =/2;
meanwhile, if /16<Re(2)</16, an approximate
integral path is taken, which is similar to Clause (b).
(d) For the condition of 0<2
When the starting point 2 is located in this region, the numerical search of the steepest descent line
becomes necessary near =/2. However, when
<Re(2)<, an approximate integral path is taken
too, which is similar to Clause (c).
The numerical search algorithm of the steepest
descent line has been discussed by Iwashita and
Ohkusu (1989) and Du and Wu (1998). When |Z| and
|Y| are close to zero compared to |X|, then the integrands of Eq. (2a) are so large at the start point and
decrease too rapidly for numerical integration to be
appropriate. For this case, a complementary scheme
should be introduced as illustrated in (Iwashita and
Ohkusu, 1989; 1992; Du, 1996; Du and Wu, 1998).

115

4.4 Self-adaptive integration method in parametric domain

The self-adaptive integration method to calculate
the integral along the steepest descent line was firstly
established by Du (1996) to evaluate translatingpulsating Greens function in Bessho form. To

Fig. 6 Arrangement of the integral path: (a)  E

<F<G<2

116

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

improve the efficiency of this method, some modifications are introduced as follows.
1. The steepest descent line generally forms a
curved line of sometimes small curvature, so this line
can be represented by a typical B-spline curve. The
knots vectors are established via accumulative chord
length method so as to acquire a parameter interval
[0, 1]. The discrete points on the steepest descent line
are represented by the B-spline, which is expressed by
n

PN

C (t )

i,k

length of each segment is 1/l and the total integral is

the sum of them. Here, the high order trapezoidal
integration rule is used and, if necessary, repeatedly
subdivides the range automatically. Details are listed
by Du (1996) and Du and Wu (1998).
I (t1 , t2 )

i 0

where Pi (i 0, 1, ..., n) is a set of control points coinciding with the coordinate of the steepest descent
line. N i , k (t ) is the basis function of the B-spline,
which can be defined as follows:
N i ,0 (t )

N i , k (t )

1, ui d t d ui 1 ,

0, otherwise,

(t  ui ) N i , k 1 (t )
ui  k  ui

(19)

(ui  k 1  t ) N i 1, k 1 (t )
ui  k 1  ui 1

(20)

uk d t d uk 1 ,

t2

t1

f (t )dt

1
| ('t )[ f (t1 )  f (t2 )]
2
1
 ('t ) 2 [ f c(t1 )  f c(t2 )]
4
1
 ('t )3 [ f cc(t1 )  f cc(t2 )].
12

(18)

(t ),

(22)

The self-adaptive integration method presented

here is an integral method performed one time along
the steepest descent line represented by the B-Spline
curve. The refined integration points are just on the
B-spline curve, without any discretization of the integration relating to the discrete path. This avoids the
waste of CPU time for numerical search of the refined
points on the steepest descent line as obtained by Du
and Wu (1998).
4.5 Truncation of numerical integration

where ui is a knot value, U [u0 , u1 , ..., un  k 1 ] is a

knot vector. The accumulative chord length method is
discussed as follows.
If B j ( j 1, 2, ..., s ) is a discrete point on the
steepest descent line, S is the total length of this line, s
is the total number of discrete points, then:
S

s 1

BB
j

j 1

(21a)

If T o S / 2, k1Z o f, the integrands tend to

zero, so we truncate the numerical integration
somewhere on the integral path for high efficiency. It
can be known from (Du, 1996) that the truncation
error is lower than 105 when choosing a truncation
point at Re(k1Z ) d 20. The finite upper limit of the
integral is calculated by Eqs. (23a) and (23b) when
is on the real axis.

j 1

ui

0, i d k  1,

ui 1  B j B j 1
, i ! k  1, j

i  k  1.

(21b)

When the steepest descent line is represented by

the B-spline curve, one-one correspondence exists
between discrete points and parameter t, where
t[0, 1].
2. As in the self-adaptive algorithm, we first
subdivide the entire integral domain of t into several
equal segments and treat each part separately. The

m
arc cos
,
2
(m  W )

(23a)

m  20 / Z  W , T [S, S / 2],

(23b)

20 / Z  W , T [S / 2,0] [0, S / 2].

m

5 Numerical results and analysis

To check the effectiveness of the proposed numerical integration methods, curves G* and Gx* are

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

plotted in Fig. 7 and Fig. 8. The same cases have been

presented by Du and Wu (1998) and Du et al. (2005).
The calculated conditions for Fig. 7 are e=1.4 rad/s,
and U=1.1 m/s, 1.5 m/s, and 2.0 m/s. Fig. 8 shows that
the source point is located at (0, 0, 0), and the field
points describe the x-axis, corresponding to Y=0. The
excellent level of agreement with known results shows
that the present numerical method is reliable.
The total average CPU time consumed to calculate G* and its partial derivatives is scattered from

4103 s to 6103 s, meeting the requirements of

engineering applications.

6 Conclusions
In this paper, the performance of Bessho form
translating-pulsating source Greens function is discussed and a numerical integration method is proposed. Conclusions can be drawn as follows.

U=1.5 (present)
U=1.5 (Du and Wu, 1998)

U=2.0 (present)
U=2.0 (Du and Wu, 1998)

(a)

(b)

Im(4SG*)

Re(4SG*)

-1

-1

-10

-5

0
r

-10

10

-5

10

4
(c)

(d)

Im(4SGx*)

Re(4SGx*)

U=1.1 (present)
U=1.1 (Du and Wu, 1998)

-2

117

0
-1

1
0
-1
-2

-2
-10

-5

-3

10

-10

-5

0
r

10

Fig. 7 Images of G* and its partial derivative for Re(4G*) (a), Im(4G*) (b), Re(4Gx*) (c), and Im(4Gx*) (d)
600

300

150
0
-150
-0.8

-1x104
-2x104
-3x104

-0.6

-0.4

-0.2
x

(b)

1x104
Re(4SG*x)

Re(4SG*)

450

2x104

(a)

z=-0.1000 (present)
z=-0.0130 (present)
z=-0.0025 (present)
z=-0.1000 (Du et al., 2005)
z=-0.0130 (Du et al., 2005)
z=-0.0025 (Du et al., 2005)

0.0

0.2

-4x104
-0.8

z=-0.1000 (present)
z=-0.0130 (present)
z=-0.0025 (present)
z=-0.1000 (Du et al., 2005)
z=-0.0130 (Du et al., 2005)
z=-0.0025 (Du et al., 2005)

-0.6

-0.4

-0.2

Fig. 8 Images of G* and its partial derivative for Re(4G*) (a) and Re(4Gx*) (b)

0.0

0.2

118

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

1. 1 is an infinite jump discontinuous node and

can be eliminated by use of the Gaussian quadrature
rule. For this rule, the end points of the integration
interval must satisfy

Im(k j Z

T T1

)  Im(k j Z

T Td

 S / 10.
2. The performances of H1 and H2 are influenced
by and Z. As for H2, only when the parameter is
very large and the value of |Z| is small, can H2 act as
high-frequency oscillatory; H1 becomes highfrequency oscillatory not only when is very large
but also when /2. However, H1 attenuates very
fast when approaches /2 since |k1Z| is also very
large and H1 becomes zero when /2. To efficiently evaluate these terms, the combination of
SDIM in the -plane and VSM can be adopted.
3. Numerical results have proved the efficiency
and accuracy of present method, and it can be further
developed to calculate the diffraction-radiation
problems of floating structures advancing in waves.

Acknowledgements
The authors would like to thank Prof. Gerard DELHOMMEAU from Laboratoire de Mecanique des Fluides,
Ecole Centrale, France, and Prof. Hidetsugu IWASHITA from
Department of Energy and Environmental Engineering, Faculty of Engineering, Hiroshima University, Japan for the
evaluation discussion on the Greens function.

References
Abramowitz, M., Stegun, I.A., 1964. Handbook of Mathematical Functions. Dover Publications, New York.
Bailey, P.A., Hudson, D.A., Price, W.G., 2001. Comparisons
between theory and experiment in a seakeeping validation
study. Transactions of the Royal Institution of Naval Architects, 143:44-77.
Bessho, M., 1977. On the fundamental singularity in the theory
of ship motion in a seaway. Memoirs of the Defense
Academy Japan, 18(3):95-105.
Brument, A., Delhommeau, G., 1997. Evaluation numerique de
la function de Green de la tenue a la mer avec vitesse da
vance. Journees de Lhydrodynamique, 25:147-160 (in
French).
Chen, X.B., Wu, G.X., 2001. On the singular and highly oscillatory properties of the Green function for ship motions.
Journal
of
Fluid
Mechanics,
445(1):77-91.
[doi:10.1017/S0022112001005481]
Du, S.X., 1996. A Complete Frequency Domain Analysis
Method of Linear Three-dimensional Hydro-elastic Responses of Floating Structures Travelling in Waves. PhD

Thesis, China Ship Scientific Research Center, Wuxi,

China (in Chinese).
Du, S.X., Wu, Y.S., 1998. A fast evaluation method for Bessho
form translating-pulsating source Greens function.
Shipbuilding of China, 2:40-48 (in Chinese).
Du, S.X., Hudson, D.A., Price, W.G., et al., 2005. Prediction of
three-dimensional seakeeping characteristics of fast hull
forms: influence of the line integral terms. International
Conference on Fast Sea Transportation, Petersburg, Russia, p.1-8.
Du, S.X., Hudson, D.A., Price, W.G., 2012. An investigation
into irregular frequencies in hydrodynamic analysis of
vessels with a zero or forward speed. Journal of Engineering for the Maritime Environment, 226(2):83-102.
[doi:10.1177/1475090211432078]
Fang, M.C., Chan, H.S., 2002. Numerical study of hydrodynamic pressure on a high speed displacement ship in
oblique waves. Bulletin of the College of Engineering,
NTU, 86:15-22.
Haskind, M.D., 1953. The hydrodynamic theory of ship oscillations in rolling and pitching. Technical Research
Bulletin, 1(12):45-60.
Iwashita, H., 1997. An improvement computation method of
the seakeeping Green function. Technical Report, the
University of Hamburg.
Iwashita, H., Ohkusu, M., 1989. Hydrodynamic forces on a
ship moving with forward speed in waves. Journal of
Society of Naval Architects of Japan, 166:187-205 (in
Japanese). [doi:10.2534/jjasnaoe1968.1989.166_187]
Iwashita, H., Ohkusu, M., 1992. The Green function method
for ship motions at forward speed. Ship Technology Research, 39:3-21.
Maury, C., 2001. Etude du Problem de la Tenue a la Mer Avec
Vitesse Davance Quelconque par une Method de Singularites de Kelvin. PhD Thesis, Laboratoire de Mecanique des Fluids, Nantes, France (in French).
Maury, C., Delhommeau, G., Ba, M., 2003. Comparison between numerical computations and experiments for seakeeping on ship models with forward speed. Journal of
Ship Research, 47:347-364.
Miao, G.P., Liu, Y.Z., Yang, Q.Z., et al., 1995. On the 3-D
pulsating source of Michells type with forward speed.
Shipbuilding of China, 4:1-11 (in Chinese).
Rahman, M., 1990. Three dimensional Greens function for
ship motion at forward speed. International Journal of
Mathematics and Mathematical Sciences, 13(3):579-590.
[doi:10.1155/S0161171290000813]
Xu, Y., Dong, W.C., 2011. A fast integration method and
characteristics research for 3-D translating-pulsating
Greens function of Havelock form in deep water. China
Ocean Engineering, 25(3):365-380. [doi:10.1007/
s13344-011-0031-8]
Yao, C.B., Dong, W.C., 2013. Study on fast integration method

Yao et al. / J Zhejiang Univ-Sci A (Appl Phys & Eng) 2014 15(2):108-119

for Bessho form translating-pulsating source Greens

function. Technical Report 2013051, Naval University of
Engineering, China (in Chinese).
Zakaria, N.M.G., 2006. A Design Aspect on the Seakeeping
Performances of Ships by 3-D Greens Function Method
with Forward Speed. PhD Thesis, Yokohama National
University, Yokohama, Japan.
Zakaria, N.M.G., 2009. Effect of ship size, forward speed and

119

wave direction on relative wave height of container ships

in rough seas. The Journal of the Institution of Engineers,
72(3):21-34.
Zong, Z., Qian, K., 2012. A three-dimensional moving and
pulsating source method for calculating motions of trimaran hull forms. Journal of Ship Mechanics, 16(11):
1239-1247 (in Chinese).


Bessho
A fast integration method for translating-pulsating source Greens function in Bessho form
Bessho


Gaussian quadrature rule
6


Gaussian quadrature rule

4103 s6103 s