|
Items in cart [1] |
Questions? Call 888-624-8373 |
| Copyright © 2009. National Academy of Sciences. All rights reserved. Terms of Use and Privacy Statement |
Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 360
A Spectral-Shell Solution for
Viscous Wave-Body Interactions
R. W. Yeung and ]. A. Hamilton
(University of California at Berkeley)
ABSTRACT
To solve viscous wave-body interaction problems, a
domain decomposition technique is used in which a
viscous flow in the near field is matched to an outer
inviscid flow. A pseudo-spectral solution of the inte-
gral equation describing the outer flow is formulated
in such a way that the outer solution can be applied
to a variety of interior-specific inner problems. This
outer flow effectively absorbs outgoing waves and can
independently supply incident waves. Application of
this outer flow to viscous and inviscid interior flows is
demonstrated and the specifics and limitations of the
matching scheme are described. The interior prob-
lems are solved by a pseudo-spectral finite-difference
technique (Yeung and Yu, 1994) which requires the
geometry to be cylindrical but does not restrict in any
way the interior and exterior flow to be axisymmetric.
INTRODUCTION
Unsteady wave-body interaction problems are typ-
ically solved using a velocity-potential formulation
that neglects viscosity of the fluid. The inclusion of
the effects of viscosity in these problems has always
been a challenge because of the extensive computa-
tional requirements associated with solving a viscous-
flow problem over the large domain required to cap-
ture wave effects, e.g., see Alessandrini and Delhom-
meau (1996) and Chen and Huang (1998~.
This paper describes a formal methodology of
matching a near-field solution of a viscous wave-body
interaction problem to the solution of an inviscid
wave-like flow in the outer-field. The advantages of
this domain decomposition method are particularly
evident in the case of zero forward speed wave-body
interactions. In these problems, vorticity is generated
primarily by the body without a substantial amount
of convective transport. The wave energy, however,
travels relatively fast and if not accounted for, reflects
off any domain-truncation boundary. This was in fact
the restriction found by the very accurate viscous-
flow work of Yeung and Yu (1994~.
The domain of the inviscid solution is assumed to
lie outside of a cylindrical "shell" which extends from
the bottom to the free-surface (see figure 1~. The
axisymmetric geometry does not restrict the general
three-dimensionality of the problem and it is helpful
to imagine this cylindrical surface as a flexible wave-
maker whose motion couples the inner and outer so-
lutions. This outer solution alone is formulated in
terms of a time-dependent integral equation utilizing
the unsteady free-surface Green function. To solve
the necessary boundary integral equation, a pseudo-
spectral technique is employed in which orthogonal
basis functions are used to discretize the solution.
Such use of the "shell-function method" (Hamilton
and Yeung, 1997), which was carried out for a sub-
merged spherical shell, allows the solution to be char-
acterized by a set of universal coefficients that are
integrals in time and space of the Green function.
These integrals are typically computationally ex-
pensive to evaluate but the methodology allows their
pre-computation and subsequent usage for any prob-
lem inside the shell. Additionally, with the axisym-
metric shell geometry, the integration and evaluation
of the free-surface Green functions are carried out us-
ing semi-analytical techniques that take advantage of
the orthogonal properties of the basis functions. The
resulting generalized solution can be applied as an
outer boundary condition for a variety of problems
and solution methods in the inner region. Boundary-
integral methods have previously been used in the in-
ner region to solve nonlinear body motion problems
(Hamilton and Yeung, 2000~.
Even if one were to solve a purely potential-flow
problem in the time domain, the outer boundary con-
dition at an arbitrary truncation surface cannot be
OCR for page 361
stated as a simple condition since generated waves points on the bounding surface. This global property
are dispersive and there are memory effects relating cannot be ignored because the outer field, despite the
the current solution to the past solution. Difficul- presence of waves, is elliptic in property. A predictor-
ties encountered in the early treatments of the open- corrector scheme is thus needed to find the solution
boundary condition that are "non-reflective" can be on the matching surface that satisfies the boundary
found in Yeung (1982a). A condition relating the conditions on the inner and outer flow-fields simul-
velocity potential ~ and the normal derivative of po- taneously. This new treatment is described and its
tential ~~ in a global (rather than point-wise) sense
was first used by Nestegard and Sclavounos (1984)
for water-wave problems in the frequency domain.
Global interaction means all points on a matching
surface are interacting, resulting in a matrix rela-
tion. The ~ to (Pn relation was used simultaneously
in three-dimensional time-dependent problems by Lee
(1985) and Lin et al. (1985), the former used what
we now call a "shell", the latter used a vertical line-
dipole, to represent the outer flow. In a review pa-
per, Yeung (1985) introduced the concept of "shell
functions" to be used in conjunction with a "general
shell." The work defined how a set of universal coe~-
cients could be computed only once and re-used sub-
sequently for any three-dimensional time-dependent
interior problems. The idea was pursued further by
Yeung and Cermelli (1993) and Hamilton and Ye-
ung (1997, 2000) for a submerged shell with success
in two and three dimensions, respectively. In a re-
lated context, Keller and Givoli (1989) conducted
stability studies of the Dirichlet-to-Neumann (DIN)
formulation for acoustic problems. However, only
recently has the shell-method's effectiveness been
fully realized, particularly with the use of a pseudo-
spectral outer representation and the introduction of
a surface-piercing shell. Lin et al. (1999) reported
computations for ship-motion problems using similar
ideas, but provided no details on the treatment of
the matching procedure, particularly some of the dif-
ficulties associated with "surface-piercing" matching
surfaces.
In this paper, the exterior solution provided by
the shell-function method is used as an outer bound-
ary condition for two types of internal problems, in-
viscid and viscous fluid. Inviscid internal problems
have been applied to this type of domain decom-
position before (Lee, 1985; Dommermuth and Yue,
1987), primarily for the case where the interior prob-
lem is also solved by a boundary-integral equation
method. The matching to a finite-difference solu-
tion of the interior problem presents a new challenge.
Finite-difference methods require point-wise bound-
ary conditions while the integral equation solution
of the outer flow provides only global condition. As
explained earlier, a global condition is a relation in
which the flow characteristics at one point are de-
scribed in terms the flow characteristics at all other
2
accuracy verified here. Somewhat surprisingly, when
tested in the frequency domain, the technique de-
scribed here does not seem to demonstrate the usual
irregular frequency effects associated with the use of
a free-surface Green function.
The predictor-corrector matching scheme can be
extended to provide an outer boundary condition for
the solution of viscous flow in the interior region.
In this work, the viscous problem in the interior re-
gion is solved using a spectral Navier-Stokes equa-
tion solver for cylindrical geometries, developed by
Yeung and Yu (1994) and previously presented for
the case of a closed domain in the 20th ONR Sym-
posium. The inviscid-viscous matching technique is
applied to two types of scenarios: (a) an initial-value
problem with waves in the viscous region passing out
through the matching boundary without reflection,
and (b) a transient-wave problem with waves, exist-
ing initially in the outer inviscid region, propagating
into the viscous interior region and diffracting about
a body.
INVISCID OUTER SOLUTION
In this section, a general solution of the linear, invis-
cid, outer flow is found by solving a boundary-integral
equation for a velocity potential. A representation
of this potential in terms of Chebyshev-Fourier basis
functions allows the general solution to be expressed
in terms of integrals of the free-surface Green func-
tion. Numerical evaluation of these coefficients is per-
formed differently than those in classical boundary-
element method applications. The semi-analytical
integration methods used are described below. The
goal of this outer solution is to provide a general solu-
tion to the outer flow which can be used as a bound-
ary condition for a variety of interior flow problems.
This work is a "'surface piercing" extension of con-
cepts demonstrated in Hamilton and Yeung (1997~.
Mathematical Problem for Outer Flow
As illustrated in figure 1, the outer fluid domain VO
is bounded by the matching surface SS, the free sur-
face SFO' the bottom SHOP and a far-field surface A,
located at infinity. All quantities in the problem are
non-dimensionalized by a characteristic length L, a
characteristic velocity U. and fluid density p, where
OCR for page 362
Outer h
Flow
z, ,
a: ,~6 '-) x
So
·P = (r'E'z')
Q = (a, 8.Z)
~ss
Figure 1: Schematic of shell geometry.
the tilde indicates dimensional variables. For each
~
problem studied, U and L can be chosen as appropri-
ate and will be specified as needed.
The velocity of the fluid is given by the gradient of
a velocity potential that is initially zero and satisfies
the Laplace equation at all time in the fluid domain,
V2~(P,t) = 0 P ~ Vo (1)
Linearized conditions are applied on all boundaries of
the fluid domain,
ptt (P. t) + (LIZ (P. t) = 0 P ~ SFO (2)
/)Z(P, t) = 0 P ~ SHO (3)
AMP, t) + B~l/(P, t) = F(P, t) P ~ SS (4)
where the subscript I' indicates differentiation with
respect to the surface normal, directed in the positive
r direction on SS, and U is taken as A. The con-
stants A, B. and the function F(P, t) are considered
known, their exact form depends on the particular
interior problem. Additionally, initial values of sb(P)
and ¢~(P) must be provided and in the far field (as
P- ~ x), the fluid velocities must vanish.
This unsteady formulation is an initial-value prob-
lem in which the boundary conditions advance the so-
lution in time and a boundary-value problem is solved
at each time-step. This requires initial conditions on
the potential and on the free-surface elevation.
Integral Equation Solution
The hydrodynamic problem defined above is best
solved by using a Green function that satisfies some
of the boundary conditions to convert equation (1) to
an integral equation, (Wehausen and Laitone, 1960~.
An unsteady Green function G(P; Q,t—r) is used
which represents the velocity potential at time t and
position P due to the introduction of a source at point
Q. in the presence of a free surface, at time a. The
Green function must satisfy the following field equa-
tion, boundary conditions, and initial condition.
V G(P, Q. t - 7) = 5(P—Q) P ~ Vo
Gtt(P,Q,t - T) + Gz(P'Q't - T) = 0 P ~ SFO
Gz(P'Q,t-~) = 0 P ~ SHO
Additionally, homogeneous initial conditions for G
and Go must be satisfied on the free surface and the
velocity resulting from the source must vanish in the
far field. Such a function is given by Wehausen and
Laitone (1960) as the sum of a singular part and a
regular time-dependent part
G(P, Q. t—'a) = Or (P. Q) + H(P, Q. t—r) t ~ ~
(6)
By applying Green's theorem to the time derivative
of the potential and the time-dependent Green func-
tion, and integrating with respect to time, a time-
dependent integral equation can be formulated, see
Yeung (1982b). Because the Green function satisfies
the linearized free-surface boundary conditions, the
bottom boundary condition, and the far-field condi-
tions, only integrals over the matching surface remain
in the integral equation. If the geometry of Ss is
taken as a cylinder of radius a and depth h (as shown
in figure 1) and the variable of surface integration Q
is expressed in cylindrical coordinates, Qtr = a ~ z'
the integral equation can be shown to be,
- 2~(P, t)+
`~27r `~0
at J ~~e z t)Gz,fP;6,z,t=0)dzdb—
O —h
All ~ '.27r `~0 -
~ at ~ ~6 z T)H~(P;6,Z, t—r~dzdd d7=
O O —h
'.27r `~0
at ~ ~ ¢§ z t)G(P;6,z,t=0)dzdd—
O —h
fit - rem r°
a: ~ (PIJ(§ z 7)HT(P;6,Z, t—7)dzd~ dr
O O —h
(7)
Where the field point P lies on the cylinder and
H(P; Q. t—irk is the time-dependent part of the free-
surface Green function, G(P; Q. t—r).
Time Integrations
The integrations with respect to time in equation (7)
represent a convolution in time of the solution and the
free surface Green function. The initial step in per-
forming these integrations is an integration by parts.
3
OCR for page 363
As an example, consider the last time integral in (7~. where,
I =
Jt - t2~ to
aJ ~ (P (A z T)HT(P;6,Z, t—r)dzdb do
O O —h
- 27r 0 t
= a | | it z T)H(P; A, Z. t—T)
O —h O
t2~ r° pt
at ~ ~ (PI7(6 z T)H(P;6,Z, t-'r)d~dzdb
O —h O
(8)
To perform these integrations in time, the solution up
to the current instant of time to is discretized by the
time sequence Elk = i(tk), (tk = kAt; k = 0, . . ., K),
and the potential is assumed to vary in a prescribed
way between time-steps. A linear variation is used
here, which removes ~~ from the time integrals, leav-
ing only time integrals of the Green function which
can be evaluated analytically. This formulation shows
clearly how a higher order variation (quadratic, cu-
bic, etc.) of potential between time-steps can be im-
plemented. Previous work of Young (1982b) did not
utilize this integration by parts but rather assumed
the potential to be constant and equal to the average
between time-steps. That constant-potential formu-
lation is the lowest order approximation possible and
can be recovered from (8) by recognizing the contri-
bution from the infinite rate of change in potential
due to the jumps at the beginning and end of each
time-step. Each higher order of approximation intro-
duces a new term into the integral equation which is
another time integral of the Green function. Because
this outer solution is developed for a fixed cylindrical
geometry and a "compute once, use many times" ap-
proach is being used, this is not a penalty as it would
be in the case where the boundary integral equation
method is applied directly to a moving body.
With the approximation of linearly varying poten-
tial, ~,~ is removed from the time integrals and the
remaining time integrals of H can be defined as:
~K,k~p;§ z) = ~ | H(P;6,z, tK—T)dr (9)
The first integral on the right hand side of (8) is zero
because the potential at time zero is vanishes and
H(P;0,z,O) = 0. The integration in time may be
written as a sum,
K—1 27r 0
I ~ Am, a / / A, Z,tk) INK kelp; 6, z~dzd§-
~ (P; §' Z) = {>K k+1 INK k k K
By defining ~K-k~p; A, Z) as an analog to
INK k ~ p; a, Z) with Ho (P; 6, z, tK—T) in place of H.
the integral equation (7) for the potential outside of
the cylinder at time to can be written in terms of
these "shell functions":
—2~r~ (P. tK ~ +
r2~ r0
a; ~ ale, z, tic) [GT/(P; §, z, O)+~°(P; 6, z)]dzd0-
O —h
r2~ r0
a) ~ ~,,(6, z, tu) [G(P; §, z, O) +~°(P; 6, z)] dzdb =
O —h
K—1 27r 0
~ al | (p(0 t ~ K—kelp ~ id do
k=1 0 —h
K—1 27r 0
~ a; | BUNS z tk)AK k(P;§ z~dzd~ (12)
k=1 0 —h
The time-discretization procedure has split the in-
tegral equation into terms that involve the velocity
potential at the current time to and terms that in-
volve the velocity potential at past time-steps. These
"memory" terms have been collected on the right
hand side of equation (12) as known quantities.
Pseudo-Spectral Solution of Integral Equation
To solve the integral equation (12) numerically, it is
necessary to represent the solution discretely, enforce
the integral equation at a set of collocation points,
and solve the resulting linear system. In this work,
the geometry of the integration surface is exploited by
representing the potential over the cylindrical shell
in terms of global orthogonal functions, Chebyshev
polynomials [Tj~x) = cos(j cos-1 x)] are used in the
vertical direction and Fourier components are used
in the circumferential direction. This choice of global
basis functions is a deviation from the usual piece-
wise approximation of earlier works (Hamilton and
Young, 1997) and superior accuracy of the solution
can be demonstrated.
The representation of ~ on Ss by orthogonal poly-
nomials is summarized by the following equations
J—1 N/2 - 1
(~(§,z,t) = ~ ~ ~nj~t)T2j~h + Lena (13)
k—1 JO J—h j=o n=—N/2
27r 0 J—1 N/2 - 1
a; | cute z tu'>°(P;e zy~z(lc (joy (p~(§,z,l) = it, ~ ~ovni(~)T2j(—h + 1)e (14)
O —h j=o n=—N/2
4
OCR for page 364
OCR for page 366
OCR for page 368
OCR for page 369
OCR for page 370
OCR for page 371
OCR for page 372
OCR for page 373
OCR for page 374
OCR for page 375
Representative terms from entire chapter:
green function
The prime on the first summation indicates that a
factor of one half should be included when j = 0. In-
serting these decompositions into the integral equa-
tion removes ~ from inside the integrals and the ex-
pansion coefficients cPnj become the unknowns. The
integrals that remain in (12) are defined as Fourier-
Chebyshev coefficients of the Green functions, h
~ ~ ~ l;2~/~0G¢P;~9 z O>~T2j(~h + 1~einedzd~
(~15~)
r2~r r °
j( ) 2~Jo J_GL,(P;~,z'O)T2j(h + l)ein~dzdd
(16)
1 r27r rO
to J ~ (P; 0, Z)T2j(h + 1)
(17)
r27r r°
nj (P)= 2~) / BY (P; §' Z)T2j(h + l)
(18)
A A
Note here that G and G`, are coefficients of the im-
pulsive (or Rankine) part of the Green function (i.e.
the Green function evaluated at t = 0) and fly and ~
are coefficients related to the time-dependent part of
the Green function as defined by equations (9) and
(11). Denoting sonKj = inj(tK) and inserting these co-
efficients into the integral equation (12), one obtains
the following integral equation for ~ at time to,
2a
a
it=
' ~ Q(a,O,z)
p(r',~',z')
: ~
\ /
R=2asin 1 2 1 (ifr' = a)
Figure 2: Polar coordinate system for evaluation of
Green function coefficients on shell surface.
COMPUTATION OF SHELL COEFFI-
CIENTS
Because the formulation presented here is intended to
be a universal solution of the outer flow that can be
applied to a variety of internal problems, it is impor-
tant to determine the most efficient way to compute
and store the various coefficients for re-use.
Regardless of the Green function in question, when
P and Q are expressed in cylindrical coordinates, the
coefficients (15) - (17) are defined by integrals of the
form,
1 r27r Ib0
2~1o ;-h
T2j~h + 1)dzein~d~
~—1 N/2 - 1
(p(P,lKi)+aE, A, (pnj (Gv~~j(~P)+7nj(~P)] =e ins Gnj~z Are) (21)
j=0 n=—N/2
J—1 N/2 - 1
a ~ ~ (PVK j tGnj~p) + ~nj(P)] =
j=0 n=—N/2
K—1 "T_1 N/2 - 1
a ~ ~ ~ Nicks—kelps_
Thus, all required coefficients can be found easily
from evaluations of Gnj (z', r', 0).
The unsteady free-surface Green function that sat-
isfies all boundary and initial conditions in (5) is given
by Wehausen and Laitone (1960),
k=l j=0 n=—N/2
K-l J-l N/2-l G(
H is given by,
H(P; Q. t—7) =
J2 °° cosh k(z' + h) cosh k(z + h)
0 sinh kh cosh kh
|1—cosELw(k, h)(t—a)] Jo(kR)dk (24)
the Fourier series representing the singular part is
needed. This series is found analytically to be,
lntR] = In [2a sin 2~] = lnta] + ~ k costly (27)
Although non-singular, the time-dependent part of
G is expensive to compute for a different reason. The
last integrand in (25) is highly oscillatory with in-
creasing k. The spatial integrations in (18) and (17)
can be performed analytically in the ~ direction with
the help of the 'addition formula' for the Bessel func-
tion,
where w(k, h) = Ok tanh kh.
Although efficient methods have been developed
to evaluate the unsteady free-surface Green function
(Beck and Liapis, 1987; Newman, 1985), the total
computation is still intensive in panel methods be-
cause the integrations in space over the elements are
typically done numerically. This can require many Jo(kR)= Jo(kr'jJO(ka)+
evaluations of the Green function, especially at large
times when the Green function varies rapidly. Be-
cause the shell surface is a fixed cylindrical surface,
and global basis functions are used on this surface,
analytic integrations can be employed to efficiently
and accurately compute the shell coefficients.
Integrals of (24) with respect to ~ are needed in
the formulation. This integration is performed ana-
lytically,
Jtk
Auk—k = H(P; Q. tK—T)d'T =
tk—1
cosh Liz' + h) cosh k'z + h
2(tk—tk_1 ) L sinh kh cosh kh Jo (kR)dk
Too cosh k(z' + h) cosh k(z + h)
+2 ~
Jo w (k, h) sinh kh cosh kh
tk
sin[w(k, h) (tK—T)] Jo (kR)dk (25,
tk—1
Integrals in space of these Green functions need to
be evaluated as defined by equations (15) - (17) and
figure 2. The singular terms in the Rankine part of
the Green function (Gr) are integratable but direct
numerical integration fails due to the singularity. To
overcome this, the singular part must be subtracted
and accounted for analytically. For the Rankine part,
the technique developed here is to perform the inte-
gration in the vertical direction first, resulting in a
function G. of §. The remaining integration with
respect to ~ is equivalent to finding the Fourier coef-
ficients. However, Gas is singular at ~ = 0 (R = 0~.
The strength of the singularity can be shown to be,
—2T2j(h +l~lntR] (26,
Subtracting this singular part from Gil results in a
regular function, the Fourier coefficients of which can
be found by FFT. To complete the evaluation of G',
oo
2 ~ Jm (kr/) Jm (ka) cos(mb) (28)
m=1
Multiplication of both sides by cos rid and integration
from 0 to 2~ provide the needed integrals,
1 r2~
IJo(kR) cos ruddy = Jn(kr') Jn(ka) (29)
2~ o
The introduction of a second Bessel function in (29)
increases the rate of decay of the integrand by a factor
of 1/~ for large k, making numerical integration
more tractable. The remaining integration in the ver-
tical direction is well behaved and can be done numer-
ically with a change of variable and Filon quadrature.
APPLICATION TO INVISCID FLOWS
Matching of a linear outer-flow field has been done
before for the specialized case when the interior prob-
lem is also solved by a boundary-integral equation
method (Lee, 1985; Dommermuth and Yue, 1987;
Hamilton and Young, 1997~. In these studies the
boundary of the interior domain is discretized in the
same manner as the outer region and the relation
between ~ and A', represented by (20) is included
implicitly in the solution of the interior problem. In
this technique, the entire flow field (inner and outer)
is solved for simultaneously at each time-step.
Recently, volume-discretization methods have been
found to be competitive with the boundary-integral
methods because of the sparseness of the resulting
linear systems (Ma et al., 2001~. However, finite-
difference and finite-element methods for solving the
interior potential-flow problem require a point-wise
boundary condition in which the relation between po-
tential and normal velocity is specified at each point
on the surface. Unfortunately, the shell method for
the outer flow admits only a global relation between
6
z
~ ~ ~r-_~
C ~ ~—
' it=
Figure 3: Schematic of an inviscid interior problem.
the shell-function method, the general outer flow so-
lution from the previous section is used to provide
a boundary condition that exactly mimics the wave-
like outer flow. In the inner region, numerical inte-
gration of the free-surface boundary condition is used
to advance the potential and wave elevation from one
time-step to the next.
The hydrodynamic forces and moments on the
body are defined as the integral of pressure over the
body:
F(t) = / P(
no.
on
n1
n ?
0 20 40 60
time
Figure 4: Horizontal force on cylinder as function of
time, co = ~r/4, A = 0.05.
the new time-step and stabilizes the method. Specif-
ically, for the predictor step, we write
lit at I (40)
tK—1/2 tK—3/2
or equivalently, in differencing form:
~u = 2¢K—1 _ oK—2 (41)
which provides a new value of ~K. The outer solu-
tion represented by equation (20) now provides OK
on SS, which is used with the matching conditions as
a boundary condition for the inner flow. Note that
the elliptic property of the outer flow is preserved in
this manner. A corrector step uses the predictor-step
solution to obtain a better estimate of OK,
ant = Veil`' ~ (42)
tK—1/2 tK—1
OK = OK
This OK value is again used with equation (20) to
provide a boundary condition on the inner flow to
complete the advancement. This procedure can be
seen as using the outer solution to provide the pres-
sure on Ss (time-derivative of velocity potential), ap-
propriate to the outer flow, subject to the predicted
normal velocity on Ss.
An alternate matching procedure is to predict the
velocity potential ~ on Ss (instead of ~~) at the next
time-step and use the outer solution to find the re-
sulting normal velocity, thus providing a Neumann
boundary condition for the inner problem. This al-
ternate procedure is found to be unstable because of
l
Added Mass
Damping
2
3
5
6
7
8
Figure 5: Non-dimensionalized sway added mass and
damping as a function of wave number.
an amplification of errors near the free-surface/shell-
surface intersection. It appears that a slight error in
the advancement of ~ leads to an error in the ver-
tical velocity, which is coupled with Mitt through the
free-surface condition. Another consideration is that
equation (19) is a Fredholm integral equation of the
second kind if ~ is considered unknown instead of in,
thus offering more stability. The resulting linear sys-
tem (20) has a diagonally dominant matrix A, which
makes solving ~ from (pn more stable than solving i)n
from A. The instability associated with a Dirichlet
condition was found to be weaker if extremely small
time-steps are taken. This is the first documented
three-dimensional work involving a surface-piercing
matching that provides a consistent and perfectly
transmissive outer condition. Campana and Iafrati
(2001) reported some success of the Dirichlet match-
ing but for a two-dimensional convective flow and a
submerged configuration.
Inviscid-fluid Results
This explicit matching procedure is applied to the un-
steady swaying of a vertical cylinder in finite depth
water. Characteristic length is chosen as the radius of
the cylinder ri and U is defined as >/~. The cylin-
der is initially at rest and for t > 0 the prescribed
horizontal velocity is given by U(t) = Aw sin wt. The
potential flow in the interior region is solved by the
PSFD method. In this case, the interior region is
truncated at a radius of five times the cylinder radius
and the shell solution accounts for the wave behav-
ior outside of this region. Figure 4 shows the result-
ing horizontal force, non-dimensionalized by pgri3,
for A = 0.05 and ~ = ~/4. After a few periods of os-
8
~ = 15.0
~ ' i , ~r
-2~ ,',: :~2' ~4,~-0-1
x
t=30.0 _ , ~s
~ i__ _ _
_ __ ~
~_ _ ~
I, . _ ~
4~,,0~:
t=20.0
1 ~ ' i - ,
l W_ l
. W_ 1
y non \ ~
0.1
O ~
at, . , ~ ~-4' 'I -0.1
'to,
x
Figure 6: Snapshots of incident waves diffracting about cylinder.
cillation, the force on the cylinder achieves a steady
state, and importantly, this steady state persists for
many periods after the waves generated by the body
oscillation have passed out of the domain of the in-
terior solution, having been effectively accounted for
by the outer shell solution and the explicit matching
technique. The amplitude and phase lag (thus the
added mass and damping coefficients) of the steady-
state solution agree well with the analytical solution
(Young, 1981) in the frequency domain.
In order to verify the performance of the match-
ing technique across a range of frequencies, this sim-
ulation is repeated for many frequencies of oscilla-
tion corresponding to a non-dimensional wave num-
ber range of k = (O. 8~. At each frequency, the added-
mass and damping in sway motion is computed from
the steady state part of the force response. Figure 5
shows the computed results along with the analytical
solution of the frequency-domain problem achieved
by separation of variables. Excellent agreement is ev-
ident but more significantly, there appears to be no
irregular-frequency effects that are usually associated
with the use of a free-surface Green function. The
normal breakdown of the solution for a vertical cylin-
der with the same radius of the shell occurs at the
roots of the Bessel function of order one: Jo (krO) = 0.
A very dense grid of computations is carried out near
the first root k = 0.7663412... and no irregular be-
havior is observed. Neither is there irregular behav-
ior occurring at the first root of J~(kri) = 0 (or
k = 3.831706...), corresponding to an irregular fre-
quency based on the dimension of the physical cylin-
der (rather than the matching surface). This amazing
property of the time-dependent shell warrants further
study.
Incident-wave problems can be studied with the
shell-function method by superposing the incident
wave in the outer region only and solving for the to-
tal potential in the inner region. This departure from
the usual technique of modifying the body boundary
condition to reflect the presence of incident waves is
done with the aim of including nonlinear effects in the
inner region, providing an outer boundary condition
that not only absorbs outgoing waves but that can in-
dependently supply incoming waves. The flow about
a vertical cylinder in a transient incoming wave field
using this technique is computed and the free sur-
face elevation is shown in figure 6. To create a strong
transient effect, the incident wave is generated by the
collapse of a two-dimensional hump of fluid outside
of the shell surface. The elevation at t = 0 is given
by:
710(X)= xy~e-(X+7.5)2/2~ (44~
with ~ = 0.5. Note that transient diffracted waves
are generated and pass out of the interior domain
as the free surface returns to a quiescent state. The
dominant diffracted wave is a ring wave that is clearly
seen to be passing out of the interior region after the
incident wave has passed. The time history of the
9
Inviscid
Flow
7
Figure 7: Schematic of viscous-flow problem.
The incompressible Navier-Stokes equations with
their associated boundary conditions are solved using
x primitive variables: velocity V and pressure p. The
velocity vector in cylindrical coordinates has the com-
ponents a, v, and w in the radial r, circumferential
0, and vertical z directions, respectively. Kinematic
conditions are required on all boundaries and on the
free-surface, stress-continuity relations appropriate to
the wave behavior are also included.
The unsteady Navier-Stokes equations in cylindri-
cal coordinates are:
horizontal hydrodynamic force associated with this
transient event is shown in figure 10 and is plotted kit + us, + - ,~ + wig
with the wave slope at ~ = 0. 1 [v2u _ u _ 2 dv] _ dP
APPLICATION TO VISCOUS FLOWS
This section applies the shell solution for the outer
inviscid flow as an outer boundary condition to a
viscous-flow problem in the interior region. The
matching technique is similar to the case of an in-
viscid interior flow described above. The interior
viscous-flow problem remains challenging in three
dimensions. In the present validation, a spectral
fractional-step method for solving the Navier-Stokes
equations in cylindrical coordinates is employed.
This technique was developed by Young and Yu
(1994, 2001~. A limitation of this technique is that it
requires axisymmetric geometries, although there are
no limitations on the flow itself, as can be seen from
the more recent convective-flow computations of Ye-
ung and Yu (2001~. The high accuracy and efficiency
of the method make it attractive for developing a test
of the viscous-inviscid matching techniques presented
here.
Formulation of Viscous Inner Flow
MU MU V 8~ 811 112
TV TV V0V TV US
+U + - +W — =
At Or r 00 Liz r
Re [V v - r2 - r2 06 I
dw dw v dw bw
at +u`, + - ,~ + w .~ =
_ 1 HIP (46)
Re [V w] - ,~~ (47)
Here, the Laplacian operator in cylindrical coordi-
nates is, V2 = 02/0r2 + 0/rbr + 02/0r232 + 02/~z2.
The continuity equation in cylindrical coordinates
completes the Navier-Stokes equations.
1 0(w) + ~ i~33v + jaw = o. (48)
The quantity P used in this section, which is not to
be confused with the field point variable in the ear-
Figure 7 illustrates the physical geometry of the inner lier sections, represents the non-dimensional dynamic
flow being considered. A vertical cylindrical strut Of pressure. It is related to the total pressure p by
radius ri extends from a flat bottom to the free sur-
face. Outside of the cylinder is an annular region
filled with a viscous fluid, the outer cylinder shown
can either be a rigid wall, or the more interesting
scenario, a matching surface, on which the bound-
ary condition provided by the shell method which
behaves the same as a wave-like outer flow.
As before, all quantities in figure 7 are non-
dimensionalized by a characteristic length L, taken
as ri, a characteristic velocity U. and fluid density
p. The physical constants of viscosity and grav-
ity will appear as the non-dimensional parameters of
Ffoude number Fr = U/~, and Reynolds number
— —
Re= UL/z/.
P=p+ F2'
r
(49)
On the cylindrical body, prescribed velocities
(U(t), V(t), W(t)) are enforced:
u = U(t), v = V(t), w = W(t), at r = ri (50)
There are no limitations on V(t) and W(t), but U(t)
should be such that the resulting motion of the inner
cylinder is small enough to be well modeled by the
linearized conditions applied on r = ri.
On the free surface, the boundary conditions can
be linearized from the exact kinematic and stress-
continuity relations (see e.g., Wehausen and Laitone,
10
1960). The linearized dynamic boundary conditions sor
are
flu + bw = 0
Liz Or
p+ ~ + 2 bw
Fr2 Re Liz
Rev 1 bw
_ + __ = 0
0, atz=0 (51)
They provide the appropriate boundary conditions
for velocities and pressure on SFO- The kinematic
condition is
,~77 = w, at z = 0, (52)
which determines the free-surface elevation A. Note
the boundary conditions in equations (51) and (52)
are satisfied on the mean free-surface z = 0, in order
to be consistent with the linearization procedure of
the outer flow.
In the original work of Yeung and Yu (1994), no-
slip boundary conditions were applied to the outer
cylinder at r = rO. This wall condition effectively
limited the length of simulations and also made the
inclusion of an incident-wave field impossible. In this
work, the rigid wall condition is replaced by condi-
tions which couple the pressure and velocities to the
outer flow through the shell functions.
Two types of boundary conditions can be applied
at the bottom of the viscous flow. A no-slip wall
condition is the most realistic, as was carried out in
Yeung and Yu (2001~.
u= 0, v = 0, w= 0, at z =—h.
Alternatively, a free-slip boundary condition is useful
for alleviating the incompatibilities of the boundary
conditions at the intersection of the bottom and the
outer cylinder at r = rO
Fx = / t~rTcosy ore sin04 dS
SB
JO (27r
=—ri dz) d6Pcos0+
—h O
Ri | demo do [2,' cost—,9 sine]
= ~ ~rrzdS
SB
= R | do; d§,09W (56)
My = J [err cos ~ ore sin d] zdS
SB
loo P27r
= - ri J zdz; d§P cost+
—h O
Ri I Adz Jo do [2~ cost—~ sine]
(55)
(57)
These quantities are non-dimensionalized by pU2ri2
and pU2r3 for forces and moments respectively. Note
that Fx and My consist of two terms, one due to
pressure and the other due to viscous stresses.
Inner Viscous-Flow Solution
A time-stepping method for solution of equations (45)
- (48) is used to solve for the velocity and pressure
field in the viscous domain at a sequence of time-
steps t = knot, (k = 1, 2, . . .~. The basic method fol-
lows Chorin (1968) and is a fractional-step method
in which an intermediate velocity field is found that
satisfies the Navier-Stokes equation with the pressure
term removed. To complete each step, a pressure field
is then found that corrects the intermediate velocity
field to form the velocity and pressure field at the
new time-step. Considering a time-difference scheme
(53) between the old (K—1) and new (K) time-steps, of
equations (45~- (48~:
~ (uK _ uK—1) = ~ t_(U . V)U +—V2U~ —VP
(58)
V uK = 0 (59)
du dv where v2 is the differential operator inside the brack-
`, = O. ,~ = O. w = O. at z =—h. (54) ets of equation (45) and Q is a suitable difference
operator. To solve these equations numerically, an
intermediate velocity field u is introduced which sat-
After the hydrodnamic problem is solved for a, v, isfies the momentum equation (58) without the pres-
w, and P. the hydrodynamic forces and moments on sure terms.
the cylinder are found as the sum of the pressure
forces and viscous stresses contained in the stress ten-
11
MENU- UK-1) = 5! [ - (U V)U+ R Mu] (60)
Subtracting this from (58J, an equation for the pres- Dirichlet boundary conditions may be used in con-
sure field results. junction with the Poisson equation for pressure at
the new time-step pa.
In the case of free-surfaces, the velocities at the
new time-step are not known, Yeung and Yu devel-
oped an algorithm in which the physical velocities
and pressures from the previous (K—1) time-step
are used in the right hand side of (63~. The boundary
condition on pressure is found from the free-surface
boundary condition and a predictor-corrector scheme
in which the pressure equation is solved twice at each
time-step. The matching technique developed below
follows a similar procedure, using the velocity of the
K—1 and K—2 steps on the matching surface to
provide boundary conditions on the intermediate ve-
locity field, then using the pressure at the new time
step supplied by the outer flow as a boundary condi-
tion on the pressure equation.
,,`~`UK _ Uy = _vpK (61)
Taking the divergence of this equation, and using the
continuity equation (59), one obtains a Poisson equa-
tion for the pressure field.
v2pK = i\t YOU, (62)
Use of (62) ensures mass conservation in the numer-
ical scheme without iteration.
The algorithm to compute the velocity and pres-
sure field at the new time-step K is to solve equation
(60) first for the intermediate auxiliary velocity field,
u. This provides a right-hand side for the pressure
equation (62~. Note that this equation for pa iS a
Poisson equation in the exact form of (39) and can
be solved by the same algorithm as that used earlier.
Finally, equation (61) provides the velocity field at
the new time-step. The solution of (60) for u must
be done accurately and is accomplished by a spec-
tral collocation method which is also developed and
presented in Yeung and Yu (1994~. The high effi-
ciency and accuracy of the techniques demonstrated
in these works stems from the cylindrical geometry
and the decomposition of the solution into spectral
modes.
In the above solution algorithm for solving the vis-
cous flow field, it is necessary to provide boundary
conditions for the intermediate velocity field u =
(u, v, w). The boundary conditions for the interme-
diate velocity can be found in terms of the boundary
conditions on the physical velocity and pressure. Ye-
ung and Yu (1994) carefully develop the boundary
conditions on the intermediate velocity in terms of
physical velocities, these results may be summarized
as:
U = fl(/\t~uK,vK WE pa)
V = f2~\t,uK,vK WE pa)
- = f3~/\t, uK, vK wK pK)
with the resulting ODE's advanced by an ADI
method.
When Dirichlet conditions are applied to the ve-
locity field, these boundary conditions can be used
directly. When free-slip conditions are applied as in
the case of the bottom surface (54), derivatives of
(63) provide the needed expressions.
Boundary conditions on the pressure field must also
be specified, either Neumann boundary conditions or
12
Matching of Inner and Outer Flows
The matching of the viscous-flow solution to the
inviscid-flow shell solution can be carried out in a
manner similar to the inviscid-interior case described
above. Again, the radial velocity on the shell surface
from the interior solution is used to predict the new
radial velocity at the new time-step, which is used
as a Neumann boundary condition for the outer-flow
solution. Since a velocity potential no longer exists in
the interior domain, the new outer potential implies
a Dirichlet condition on the pressure of the interior
flow. Again a predictor-corrector sequence is required
for stability. Hence, by analogy to equation (41), the
predictor-step normal derivative of the outer poten-
tial is
OK = 2nK—~ _ ok—2 (64)
This is used as a Neumann condition for equation
(20), which provides a new outer potential ~K. Pres-
sure on the shell surface at the new time-step is ob-
tained from a backward-difference form of the lin-
<63y earized Euler integral.
_ OK _ OK—~ (65)
The quadratic terms in velocities are not needed for
reason of consistency. Application of this pressure as
a boundary condition on the interior problem pro-
vides a better estimate for the radial velocity at the
new time-step,
K K_i + UK—us 2 (66)
y o
-1
-2
-4
_5
4
3
2
Y n
-1
-2
3
4
-5—
a -2 n
4
-0.020 -0.018 -0.016 -0.014 -0.01 1 -0.009 -0.007 -0.005 -0.003 -0.001 0.001 0.004 0.006 0.008 0.010
3
2
Y O
-1
-2
-3
4 -2 0
x
2 4 -4 -2 X
2 4
Figure 8: Wave-elevation contours ~(x, y) and velocity vectors for a non-axisymmetric Cauchy-Poisson prob-
lem in a viscous fluid near a cylinder, Re = 5, 000.
13
0.2
0.1
~ o
-0.1
-0.2
o
It=7.6~
3
1 2
.......... ...............
- - -- r0 = 10 Calculation
r0 = 5 Calculation
\
, .
;~
~ ,
0.500
0.409
0.318
0.227
0.136
0.045
-0.045
-0.136
·0.227
-0.318
-0.409
-0.500
6 7r
Figure 9: Comparison of velocity vectors and ~ component of vorticity in ~ = 0 plane for a non-axisymmetric
Cauchy-Poisson problem with the shell located at r0 = 10 (top) and r0 = 5 (bottom).
which gives the final pressure boundary condition for
the interior region after (20) is used again:
K INK _ INK—~
P —
/\t
In viscous inner flow, care must be exercised in the
tangential velocities. While derivatives of the outer
potential could also be used as tangential velocity
boundary conditions for the inner flow, consistent re-
sults are achieved by implementing a free-slip veloc-
ity boundary condition on the shell surface. This ap-
proach is based on the physical idea that in a viscous-
invisid matching, the inviscid outer flow is expected
to be incapable of supporting shear stresses on the
matching surface. The predicted radial velocity could
be used as the final, required boundary condition for
the inner flow but good results with less computa-
tion are achieved by simply using the most current
time-steps radial velocity in the expressions for the
boundary conditions on the auxiliary velocity A. In
this way, the flow is driven by an applied pressure
on the shell matching surface, analogous to the free-
surface treatment used by Yeung and Yu (1994) in
the original development of the PSFD solution of the
viscous flow problem.
This matching procedure differs from that used by ~ ~ v - v
Campana and Iafrati (2001), as noted earlier, their Figure 9 shows the flow velocity and vorticity in the
approach is the opposite, using pressure from the in- ~ = 0 plane of the interior viscous flow. Evidently,
terior flow to advance the outer velocity potential
through the Euler integral. Although they demon-
strate success and good results, that approach has
`67' been found unstable when used in conjunction with
a surface-piercing shell.
Results for a Viscous Inner Flow
To demonstrate the effectiveness of the viscous-
inviscid matching procedure, a Cauchy-Poisson type
problem is solved in which a Gaussian-shaped hump
of fluid near the body is allowed to collapse, generat-
ing waves which impinge upon the cylindrical piling
and pass out of the viscous domain. The hump is
given by the following form at t = 0:
POOPS = 0.1e-2 6r
(68)
where r- is a polar coordinate system centered at
(2.3,0~.
Figure 8 shows the wave elevation at several time-
steps after the release of this hump in the viscous
region. These plots show clearly that the shell-
boundary condition at r0 = 5 is effective in absorbing
the wave energy without reflection. For a consistency
check, the problem is solved for two radial locations
of the shell matching boundary. rig = 5 and rig = 10.
14
0.15
0.1
0.05
o
-0.05
-0.1
-0.15
Horizontal Pressure Force, Re = 1000
- ' . Horizontal Viscous Force, Re = 1000
. Horizontal Force, Inviscid Flow
Slope of Incident Wave at Origin
,1:
\ 1
0 5 10 15 20 25 30 35
time
Figure 10: Incident-wave slope at origin and resulting force on a vertical cylinder.
the viscous-inviscid shell has little effect on the flow strafed and explored in this work. This is the first
details, especially close to the body. For each of these three-dimensional free-surface piercing shell which
two test cases, the characteristic velocity U is cho-
sen as/ and viscosity is chosen to give a Reynolds
number of 5, 000.
has proven to be most efficient when used in con-
junction with a pseudo-spectral representation of the
outer flow in finite water-depth. The shell is fully
The transient incident-wave computations from the transmissive in wave properties in both incoming
inviscid section are repeated here with the viscous in-
terior problem fully coupled to the inviscid outer flow,
which supplies transient incident waves. The total
wave elevations that result have a similar appearance
to the inviscid case, with a ring wave being the dom-
inant diffracted wave which is transmitted properly
outwards by the shell boundary condition. Figure 10
compares the resulting horizontal force on the cylin-
der with the inviscid-fluid computations. The pres-
sure forces resulting from the viscous flow are similar
to the inviscid results in the early part of the simu-
lation but ultimately have a smaller magnitude and
decay faster as the transient incident-wave passes. A
viscous force due to shear stress is also present but
remains only a few percent of the pressure force since
flow separation has not developed in this short time
interval. In this solution, the flow stays attached to
the cylinder. For separation to occur, the frequencies
must be much lower than those in this example, re-
quiring a non-transient incident wave field that starts
from rest and achieves a steady state wave behavior;
but this is simply a matter of having an appropriate
incident wave for the problem.
CONCLUDING REMARKS
A highly effective outer boundary condition based on
the use of shell functions for an inviscid outer flow,
as initially introduced by Yeung (1985), is demon-
15
and outgoing directions. It is characterized by a
set of pre-computed coefficients based on the time-
dependent free-surface Green function of Finkelstein
(1957~. The pre-computed coefficients can be stored
and re-used for a variety of internal wave-body inter-
action problems.
A new "explicit" point-wise matching procedure is
presented for coupling the interior solution to the
fully transmissive shell. The procedure enables the
advancement of both interior and shell solutions in a
stable manner. The basic reasoning relies on the fact
that the velocity potential in the outer field must re-
ceive a normal-velocity condition based on a matched
advancement of the pressure field on the shell. The el-
liptic property of the flow is preserved through the use
of time-dependent Green's theorem which relates the
potential and its normal derivative. This simple pro-
cedure is validated for several Cauchy-Poisson prob-
lems with the presence of a vertical cylinder as a body
in the interior region. Solutions in the interior region
are obtained by highly accurate and efficient pseudo-
spectral finite-difference (PSFD) methods for invis-
cid flow (Yu & Yeung 1995) and full Navier-Stokes
flow (Yeung & Yu, 1994~. When used in combination
with the shell theory and the new matching condi-
tion, these PSFD methods can uncover interesting
fluid physics associated with the viscous interaction
of surface waves and a body heretofore not possible.
References
Alessandrini, B. and Delhommeau, G., "A multigrid
velocity-pressure-free surface elevation fully cou-
pled solver for calculation of turbulent incompress-
ible flow around a hull," Proceedings of the 21st
ONR Symposium on Naval Hydrodynamics, liond-
heim, Norway 1996, pp. 40-55.
Beck, R. F. and Liapis, S. J., "transient motions of
floating bodies at zero forward speed," Journal of
Ship Research, 313:164-176, 1987.
Campana, E. F. and Iafrati, A., "Unsteady free sur-
face waves by domain decomposition approach,"
Proceedings of the 16th International Workshop
on Water Waves and Floating Bodies, Hiroshima,
Japan, 2001.
Chen, H. C. and Huang, E., "Validation of a chimera
rans method for transient flows induced by a full-
scale berthing ship," Proceedings of the 22nd ONR
Symposium on Naval Hydrodynamics, Washington,
D.C., 1998, pp. 948-963.
Chorin, A. J., "Numerical solution of incompressible
flow problems," Studies on Numerical Analysis, 2:
64-71, 1968.
Dommermuth, D. G. and D.K.P. Yue, D. K. P.,
"Numerical simulations of nonlinear axisymmetric
flows with a free surface," Journal of Fluid Me-
chanics, 178:195-219, 1987.
Finkelstein, A. B., "The initial value problem for
transient water waves," Communications of Pure
and Applied Mathematics, 10:511-22, 1957.
Hamilton, J. A. and Yeung, R. W., "Shell-function
solution for 3-d nonlinear body-motion problems,"
Schiffstechnik, 44:62-70, 1997.
Hamilton, J. A. and Yeung, R. W., "Motion of a
submerged body in incident waves," Proceedings of
the 15th International Workshop on Water Waves
and Floating Bodies, Ceasarea, Israel, 2000.
Keller, J. B. and Givoli, D., "Exact non-reflecting
boundary conditions," Journal of Computational
Physics, 82:172-192, 1989.
Lee, M. Y., Unsteady Fluid-Structure Interaction in
Water of Finite Depth, Ph.D. thesis, University of
California, Berkeley, 1985.
Lin, W. M., Newman, J. N., and Yue, D. K., "Nonlin-
ear forced motions of floating bodies," 15th Sym-
posium on Naval Hydrodynamics, Hamburg, Ger-
many, pp. 33-49, 1985.
Lin, W. M., S. Zhang, S., Weems, K., and Yue,
D. K. P., "A mixed source formulation for nonlin-
ear ship-motion and wave-load simulations," Pro-
ceedings 7th International Conference on Numeri-
cal Ship Hydrodynamics, Paris, Ffance, 1999.
Ma, Q. W., Wu, G. X., and Eatock Taylor, R., "Fi-
nite element simulation of fully non-linear inter-
action between vertical cylinders and steep waves.
partl: Methodology and numerical procedure," In-
ternational Journal for Numerical Methods in Flu-
ids, 3:265-285, 2001.
Nestegard, A. and P.D. Sclavounos, P., "A numer-
ical solution of deep-water wave-body problems,"
Journal of Ship Research, 1984.
Newman, J. N., "Algorithms for the free-surface
green function," Journal of Engineering Mathe-
matics, 19:57-67, 1985.
Wehausen, J. V. and Laitone, E. V. "Surface waves,"
In S. Flugge, editor, Handbuch der Physik, vol-
ume 9, pp. 446-778, Springer-Verlag, 1960.
Yeung, R. W., "Added mass and damping of a verti-
cal cylinder in finite-depth waters," Applied Ocean
Research, 33:119-133, 1981.
Yeung, R. W., "Numerical methods in free-surface
flows," Annual Review of Fluid Mechanics, 14:395-
442, 1982a.
Yeung, R. W., "liansient heaving motions of floating
cylinders," Journal of Engineering Mathematics,
16:97-119, 1982b.
Yeung, R. W., "A comparative evaluation of numer-
ical methods in free-surface flows," IUTAM Sym-
posium on Hydrodynamics of Utilization of Ocean-
Wave Energy, Lisbon, Portugal, 1985, pp. 325-356.
Yeung, R. W. and Cermelli, C., "Shell functions: A
global method for computing time-dependent free-
surface flows,", Proceedings of the 8th International
Workshop on Water Waves and Floating Bodies,
St. Johns, Newfoundland, 1993.
Yeung, R. W. and Yu, X., "Three-dimensional flow
around a surface-piercing body," Proceedings of the
20th ONR Symposium on Naval Hydrodynamics,
Santa Barbara, CA, 1994.
Yeung, R. W. and Yu, X., "Three-dimensional free-
surface flow with viscosity: A spectral solution,"
Hydrodynamics in Ship and Ocean Engineering,
RIAM, Kyushu University, pp. 87-114, April 2001.
Yu, X. and Yeung, R. W., "Interaction of tran-
sient waves with a circular surface-piercing body,"
ASME, Journal of Fluids Engineering, 117:382-
388, September 1995.
16
