Вы находитесь на странице: 1из 16

Journal of Applied Mechanics and Technical Physics, Vol. 49, No. 4, pp.

666680, 2008

ON EQUATIONS OF MOTION OF A NONLINEAR HYDROELASTIC STRUCTURE

P. I. Plotnikov and I. V. Kuznetsov UDC 532.5:539.3

Formal derivation of equations of a nonlinear hydroelastic structure, which is a volume of an ideal


incompressible uid covered by a shell, is proposed. The study is based on two assumptions. The
rst assumption implies that the energy stored in the shell is completely determined by the mean
curvature and by the elementary area. In a three-dimensional case, the energy stored in the shell
is chosen in the form of the Willmore functional. In a two-dimensional case, a more generic form
of the functional can be considered. The second assumption implies that the equations of motion
have a Hamiltonian structure and can be obtained from the Lagrangian variational principle. In a
two-dimensional case, a condition for the hydroelastic structure is derived, which relates the external
pressure and the curvature of the elastic shell.
Key words: free boundary, variational principle, ideal uid, hydroelasticity, constraint forces,
Antman equation, Bernoulli law.

1. Equations of Motion in the Lagrangian Coordinates. A long list of publications on the theory of
nonlinear hydroelasticity can be found in [1].
The following notation is needed to formulate the model of a hydroelastic structure, which was rst proposed
in [2] to describe waves on the surface of a liquid covered by an ice layer.
Let an ideal incompressible uid at the time t occupy a domain t in the Euclidean space of points x =
(x1 , x2 , x3 ) R3 . In turn, the shell thickness is assumed to be small, and its mid-surface coincides with the boundary
of the ow domain as geometric positions of points.
We consider the Lagrangian variables = (1 , 2 , 3 ) determining the positions of material particles. Actually,
the coordinate is a label of a material particle chosen more or less arbitrarily.
We assume that the points occupy a certain domain R3 with a smooth boundary . Then, the
positions of the uid points are characterized by the vector eld of displacements x(t, ) ( ), and the positions
of the shell particles are characterized by the eld of displacements y(t, ) ( ).
In the initial-boundary problems, it is convenient to consider as the positions of material points at the
time t = 0. In this case, we have 0 = and 0 = . Thus, the boundary of the ow domain and the shell admit
two presentations:
x
t : x = x(t, ), yt : y = y(t, ) for .
y
During the joint motion in the general case, the uid may separate from the shell; hence, the surfaces x t and t
may fail to coincide. This eect is called the partial lling of the cavity by the uid. In the present paper, the
y
eect is ignored, and further considerations are limited to the case with x t = t . The shell, however, may slip
with respect to the ideal incompressible uid, which means that x(t, ) = y(t, ) for .
Let us recall the basic facts from the theory of surfaces. If the surface locally admits parametrization
= (q ) = (q1 , q2 ), then the normal vector n and the elementary surface area have the following form in the
coordinates (q1 , q2 ):

Lavrentev Institute of Hydrodynamics, Siberian Division, Russian Academy of Sciences, Novosibirsk 630090;
plotnikov@hydro.nsc.ru; kuznetsov i@hydro.nsc.ru. Translated from Prikladnaya Mekhanika i Tekhnicheskaya
Fizika, Vol. 49, No. 4, pp. 174191, JulyAugust, 2008. Original article submitted July 30, 2007; revision sub-
mitted October 3, 2007.
666 0021-8944/08/4904-0666 
c 2008 Springer Science + Business Media, Inc.
 
1  
n((
q )) = q1 (
q ) q2 (q ), d = g0 dq, g0 = q1 (q ) q2 (q ).
g0
If, at each time t, the moving surface yt admits parametrization
Y (t, q ) = y(t, (q )),
then the vector of the outward normal to yt
and the elementary area are described by the formulas

1
q ))) =  y q1 Y (t, q) q2 Y (t, q ),
(t, y(t, ( dyt = gty dq,
gt
   (1)
 
gty = q1 Y (t, q ) q2 Y (t, q ).

The components of the metric tensor gij (1  i and j  2) and the components of the second quadratic form Lij
(1  i and j  2) are given by the equalities
Y1 Y2
gij = (Yi , Yj ), Lij = (, qi Yj ), Yi = qi Y , = .
|Y1 Y2 |
The doubled mean curvature H is calculated by the formula
H = g ij Lij , (2)
ij 1
where g = (gij ) .
y
Similar formulas describe the surface x
t , which coincides with t as the geometric set of points.
The motion of a nonlinear hydroelastic structure is characterized by the velocity elds
v(t, ) = t x(t, ) for , u(t, ) = t y(t, ) for ,
where v is the velocity of the uid particle and u is the velocity of the shell particle in the Lagrangian coordinates i
(i = 1, 2, 3). In addition, the motion is characterized by the density distributions in the corresponding components.
Without losing generality, we assume that the uid density equals unity. The shell bounding the uid is
compressible; therefore, it is necessary to use the formula for the density distribution in the shell. Let the density
distribution in the shell at the initial time in the Lagrangian coordinates be dened by the function 0 (). This
means that the mass of an arbitrary part of the shell A is determined by the equality

0 () d.
A

The law of conservation of mass implies the equality


 
0 () d = (t, ) dyt ,
A y(t,A)

which, in turn, yield the presentation



 dy 1 g0 (q )
t
(t, ) = 0 () = 0 () . (3)
d gty (q )
For simplicity, we assume that the mass is uniformly distributed in the shell at the initial time, i.e., (0, ) =
0 () = 1. Under these assumptions, the kinetic energy of the uid Kf and the kinetic energy of the elastic shell
Ke have the form

1
Kf = |t x(t, )|2 d,
2

 
1 1
Ke = (t, )|t y(t, )|2 dyt = |t y(t, )|2 d.
2 2
y
t

667
Let us put forward the following hypotheses.
1. Equations that describe the nonlinear hydroelastic structure form a dynamic system with a conguration
space (x(t, ), y(t, )) L2 ()3 L2 ()3 for t (0, T ).
2. The Lagrangian for the uid has the form

1
Lf = |t x(t, )|2 d.
2

3. The Lagrangian for the shell has the form

1
Le = |t y(t, )|2 d W (yt ),
2

where W (yt ) is the stored (potential) elastic energy of the shell, which is dened in the form of a surface integral
 
y 1
W (t ) = W ( gty , |H|) dyt ; (4)
2
y
t

the vector of the mean curvature H = H and the unit vector of the outward normal are described by Eqs. (1)
and (2); dyt is the elementary surface area. This model of the shell is used in the nonlinear theory of elastic shells
(see [3]). It should be noted that the functional W (yt ) depends on the choice of the Lagrangian coordinates and
changes its form if the independent variables are replaced. Thus, presentation (4) depends on the choice of the
coordinate (this issue requires careful consideration in each particular case). In the linear theory of elasticity, this
problem does not arise, because the Lagrangian coordinates in this theory are chosen uniquely as the positions of
particles in a certain unloaded state. It is of interest to consider the case where the functional of the stored energy
is a geometric invariant and does not depend on parametrization chosen. In the class of functionals of the form (4),
there exists only one geometrically invariant representative with a nontrivial dependence on the external curvature,
namely, the so-called Willmore functional (see [4]):

1
W (yt ) = |H|2 dyt .
2
y
t

The role of the Willmore functional in the elasticity theory was noted, e.g., in [3].
4. To derive the equations of motion, we need to describe all constraints imposed on the mechanical system
of motion. It is further assumed that there are two natural constrains. The rst constraint is the principle of uid
incompressibility, which is written as the equation
det D x(t, ) 1 for , (5)
where D x is the Jacobi matrix of the mapping  x(t, ). The second constraint reects the coincidence of the
uid surface and elastic shell as subsets of the Euclidean space in the course of their motion:
y
x
t = t . (6)
2. Configuration Manifold . Let us consider the hydroelastic structure as a dynamic system in a linear
space consisting of innitely dierentiable vector elds (x(), y()), where x : R3 and y : R3 . We assume
that has a Hilbertian structure L2 ()3 L2 ()3 . Under these assumptions, constraints (5) and (6) determine an
innite-dimensional conguration manifold . Having an induced metrics L2 ()3 L2 ()3 , the space is not
complete, and all further considerations have a formal character. The following lemma oers a description of the
tangential space to at the point (x, y) .
Lemma 1. The tangential space to the manifold at the point (x, y) consists of all vector elds [x(), y()],
x: R3 and y: R3 , satisfying the equalities
div (M 1 x) = 0 for ; (7)

(x(())) x(()) = (y()) y() for , (8)


where M = D x() is the Jacobi matrix of the mapping  x(), and the dieomorphism ():  is dened
by the equality
668
x(()) = y() for . (9)
Note that the vector (t x, t y) for the trajectory of the dynamic system [x(t), y(t)] belongs to the tangential
space Tan(x,y) ; hence, conditions (8) and (9) yield the relation
(t, x(t, (t, ))) t x(t, (t, )) = (t, y(t, )) t y(t, ),
where x(t, (t, )) = y(t, ) for .
Proof of Lemma 1. Let (x, y) . We assume that M () = D x() and |M | = det M = 1. Then, the
variation of the constraint equation (5) with respect to x has the form
|M | = div (M 1 x) = 0 for ,
whence there follows Eq. (7).
We introduce auxiliary functions x = x () and y = y () depending on the parameter (1, 1) and
generating the surfaces x y x y
= x () and = y (). The equality = allows the following conditions to be
imposed on the functions x and y :
x0 = x, y0 = y, x y
= .

Hence, there exists a -parametric family of mappings : such that


x ( ()) = y () for . (10)
In terms of the theory of Riemann manifolds, the functions x = x() and y = y() are called isometric immersions
(see, e.g., [5]); x and y are the rst-order innitely small bendings of the immersions x and y, respectively:
x = x + x + o(), y = y + y + o(), o()/ 0 for 0
(x and y are the rst-order bending elds). By xing an arbitrary value of and dierentiating Eq. (10)
with respect to at = 0, we obtain the relation between the bending elds
d 
D x(0 ()) () + x(0 ()) = y(). (11)
d =0


As the set { ()}(1,1) is the curve on , then (d /d)() is the tangential vector to at the point 0 ().
 =0

Hence, D x(0 ())(d /d)() is the tangential vector to x 0 at the point x(0 ()). Multiplying Eq. (11) in
=0
a scalar manner by the normal vector to x 0 at the point x(0 ()) = y(), we obtain Eq. (8). In what follows, we
denote the dieomorphism 0 : by .
As a consequence of Lemma 1, we obtain the following statement for the structure of the space (Tan(x,y) )
orthogonal to the manifold at the point (x, y). Note, by virtue of the choice of the conguration space , the
space orthogonal to Tan(x,y) at the point (x, y) consists of all vector elds [N (), L()], N : R3 and L:
R3 , satisfying the relation
 
N () x() d + L() y() d = 0 (x, y) Tan(x,y) . (12)

The vector elds N and L are called the constraint forces.
Lemma 2. For each point (x, y) , the space (Tan(x,y) ) consists of the vector elds (N , L) such that
N () = (M 1 ) p() for ,

1
L() = (p(()) + C)(y()) for .
()
Proof. Let h C0 () and div h = 0. We choose (x, y) Tan(x,y) in the following form: x() =
M ()h() and y() = 0. Hence, Eq. (12) takes the form

N () (M ()h()) d = 0.

As h is arbitrary, there exists a function p() such that


669
N () = (M )1 p(). (13)
We choose an arbitrary vector l() orthogonal to the normal vector (y()). In Eq. (12), we assume that
x() = 0 and y() = l(). All tangential vectors l obey the equality

L() l() d = 0,

which yields the presentation of the constraint force


L() = () (y()). (14)
We have to determine the form of the scalar function = (). Let k be an arbitrary solenoidal vector eld
C () and x = M ()k() for and y() = ((M )()k () (x ())) (x ()) for .
Taking into account the equality (x ()) = (y()) for and substituting presentations (13) and
(14) of the constraint forces N and L to Eq. (12), we obtain the equality
   
(M k) ((M )1 p) d + ()((M )k ) (x ) (x ) (x ) d = 0. (15)

As div k = 0, we have
 
(M k) ((M )1 p) d = p()k n d, (16)

where n() is the unit vector of the outward normal to :


 
n((q )) = q1 (
q ) q2 (
q )/ g0 (q ), g0 (q ) = |q1 (q ) q2 (q )| for q Q,
= (
q ) is local parametrization of .
Using Eq. (16), we write the rst integral in the left side of equality (15) in parametric form as
  
p()k() n() d = p(q )k(q ) n(q ) g0 (q ) dq,
Q

where p( q )), k(


q ) = p(( q ) = k(( q )), and n(
q ) = n((q )). Then, Eq. (15) acquires the form
  
q )k(
p( q ) n( q ) dq + (M ) k (x )() d = 0.
q ) g0 ( (17)
Q

To simplify the integrand of the second term in the left side of this equality, we nd the relation between the vectors
n() and (x()). Let X = X( q ) = x((q )). We recall that
 
q ))) = q1 X(
(x(( q ) q2 X(q )/ g x (q ), g x (q ) = |q1 X(q) q2 X(q )|.
Then,
3

qi X(
q) = qi k (q ) k x((q )),
k=1

which implies that


   
 q1 2   q1 1 
q1 X q2 X =  q1 1 x 2 x +  q1 3 x 1 x
q2 1 q2 2  1 q2 3 q2 1  3
 

 q1 3 
+  q1 2 x x = x x; x x; x x (q1 (q ) q2 (q )).
q2 2 q2 3  2 3 2 3 3 1 1 2

Note that


(M )1 = 2 x 3 x; 3 x 1 x; 1 x 2 x .

670
Hence, we obtain
M (q1 X(
q ) q2 X(q )) = (q1 (q ) q2 (q )),
which yields the relation between the vectors n() and (x()):

g0 (q )
M ((
q )) (x((q ))) = n((q )) .
g x (q )
In this equality, we replace parametrization  q (q), where the dieomorphism : Q Q satises the identity
((q )) = ((q )). As a result, we obtain the relation

g0 ((q))
(M ((
q ))) (x ((q ))) = n(((q ))) .
g x ((q ))
Multiplying both sides of this relation in a scalar manner by k , we obtain

g0 ((q ))
(M ) k (x ) = k(((q ))) n(((q ))) .
g x ((q ))
The second integral in the left side of Eq. (17) can be presented as an integral with respect to the parameter q:

((M ) k ) (x )() d

 
g0 ((q )) 
= k(((
q ))) n(((q ))) g0 (q ) ((q )) dq. (18)
g x ((q ))
Q

We apply the replacement of the variables  q r = (q ) in the integrand in the right side of this equality. In the
q ) takes the form (r ) = 1 (r ). As d is a geometric invariant, we obtain
new variables, the function (
  
d = g0 ( q ) dq = G0 (r ) dr, G0 (r ) = |r1 (r ) r2 (r )|.
From here and from Eq. (18), we nd that

(M ) k (x ) () d

 
g0 (r ) 
= k((r )) n((r )) G0 (r )((1 (r ))) dr . (19)
g x (r )
Q

Replacing r by 
q in Eq. (19), we write Eq. (17) in the following form:
 
  G0 (q ) 
1
k(
q ) n(
q ) g0 (
q ) p(q ) + (( (
q ))) dq = 0.
g x (q )
Q

As k is an arbitrary solenoidal vector eld, we have


  
k() n() d = k(q) n(q ) g0 (q ) dq = 0.
Q

Hence, there exists an unknown constant C such that



g x (q )
((1 (
q ))) = (p((q )) + C).
G0 (q )
In the latter equality, we apply the replacement of the variables q (q ):

g x ((q ))
((q )) = (p(((q ))) + C). (20)
G0 ((q ))
671
To nish the proof of Lemma 2, it suces to establish a relation between the fundamental forms g x and
g y . In the notation X(
q ) = x((
q )), Y (
q ) = y((q )), and ((q )) = ((q )), the condition y() = x () is
written as
Y (q ) = X((q )). (21)
By dierentiating condition (21) with respect to qi as
qi Y ( q ) 1 X((q )) + qi 2 (q )2 X((q )),
q ) = qi 1 (
we obtain
 
g y (
q ) = g x ((q )) det Dq (q ).
In the notation used, (
q ) = ((
q )), r = (
q ), and
q1 q2 = r1 r2 det Dq (q ),
which implies that
 
g0 (
q) = G0 ((q )) det Dq (q ).
Thus, we have
 
g y (q ) g x ((q ))
 = . (22)
g0 (q ) G0 ((q ))
Using Eqs. (20) and (22), we nally obtain
 
q) 
g y (  g y (q )  
q )) =
(( p(((q ))) + C = p ((q )) + C ,
g0 (q) g0 (q )
which nalizes the proof of Lemma 2.
3. Variation of the Willmore Functional. In calculating the variation, we use the results of [6]. Let
y():  y by a smooth immersion of the manifold in R3 . Let us recall the notation of parametrization of the
surface y : Y ( q )), where q = (q 1 , q 2 ). Let us consider a new immersion
q ) = y((
Y (q ) = Y (q ) + Y (q ).
We decompose Y (
q ) into the tangential and normal components:
Y =  Y + Y = i Yi + .
Then, the tangential variation of the functional W has the form
  
1 1 1 1 1
 W (y ) = y qi ( g y i H 2 ) dy = y qi ( g y i H 2 ) dy = div (H 2 ) dy ,
2 g 2 g 2
y y y

where  = (1 , 2 ). As the surface y is closed we have  W (y ) = 0.



The normal variations of H, g y , and dy are calculated by the formulas

dy = H dy , g y = H g y , H = gy + Lij Lij ,
where gy is the LaplaceBeltrami operator dened by the expression
1
gy = y qi ( g y g ij qj ).
g
The normal variation of the functional W (y ) is determined as
  
1 1
W (y ) = (H 2 dy ) = H H dy + H 2 (dy )
2 2
y y y
  
1
= H(gy + Lij Lij ) H 3 dy .
2
y
672
We simplify the expression to
  
H(gy + Lij L ) d = (gy H + Lij L H) d + qi (Hqi qi H) dy .
ij y ij y

y y y
y
As is a closed surface, we have

qi (Hqi qi H) dy = 0.
y
Hence,
  1  
W (y ) = g H + H 2 + R H dy ,
2
y
where R is the scalar curvature of the surface y satisfying the GaussCodacci equation
Lij Lij = H 2 + R.
4. Lagrangian Principle. Governing Equations of the Hydroelastic Structure. Now we can
derive equations of motion of the hydroelastic structure in the Lagrangian variables. For simplicity, we consider
the case where the stored energy of the shell is dened by the Willmore functional. The governing equations of the
hydroelastic structure with constraints can be derived with the use of the Lagrangian variational principle
Lf N (x) + Le L(y) = 0 (23)
for all smooth functions (x, y). Here the linear functionals N and L are dened by the equalities
 
N (x) = N (t, ) x(t, ) d, L(y) = L(t, ) y(t, ) d

(N and L are the constraints). Using Lemma 2 and the expressions Lf , Le , and W in explicit form, we write
Eq. (23) as
  
t2 x(t, ) + (M 1 ) p(t, ) x d

  
 1   gy
t
t2 y(t, (
q )) + gty H + 2
H +R H (t, y(t, (q )))
2 g0
Q

gty 
(p(t, (t, (
q ))) + C(t)) (t, y(t, (q ))) y g0 dq = 0.
g0
Within the framework of Hypotheses 14 (see Sec. 1), the equation in variations (23) is equivalent to the
following boundary-value problem of the dynamics of the hydroelastic structure.
Problem A. We have to nd time-dependent dieomorphisms x(t, ):  t R3 and y(t, ): 
y
t R3 , a function p(t, ):  R, a function C(t), and a family of dieomorphisms (t, ):  satisfying the
following equations:
x(t, (t, )) = y(t, ) for ; (24a)

(t, x(t, (t, ))) t x(t, (t, )) = (t, y(t, )) t y(t, ) for ; (24b)

t2 x(t, ) + (M 1 (t, )) p(t, ) = 0, det M = det D x(t, ) 1 for ; (24c)

(t, )t2 y(t, ) + (gty H + (H 2 /2 + R)H)(t, y(t, ))

= (p(t, (t, )) + C(t))(t, y(t, )) for ; (24d)

x(0, ) = for , y(0, ) = , (0, ) 1 for .


673
Here H = H(t, y(t, )) is the doubled mean curvature yt at the point y(t, ) and (t, y(t, )) is the unit vector of
the normal to yt at the point y(t, ); the density of the elastic membrane is dened by the formula [see Eq. (3)]

d g0 (q )
(t, ) = (t, y(t, )) = , (24e)
dt gty (q )
and 
q is an arbitrary local parametrization .
5. Formulation of the Problem in the Eulerian Variables. In the Eulerian formulations, the functions
depend on the Cartesian coordinates x = (x1 , x2 , x3 ) and the time variable t. Let us recall that the uid occupies
the domain t with the boundary x t at each time instant t [0, T ]. Let

QT = t {t}, ST = xt {t}.
t(0,T ) t(0,T )

The vector of the normal to ST is denoted by = (x, t), as in the Lagrangian coordinates. We use v(x, t) =
t x(t, (x, t)) to denote the uid velocity and u(y, t) = t y(t, (y, t)) to denote the membrane velocity. Let the
material surface ST of the domain occupied by the uid be dened by the equation F (x, t) = 0. Hence, the uid
velocity satises the kinematic condition
t F (x, t) + v x F (x, t) = 0 for F (x, t) = 0.
In the Eulerian formulation, it makes no sense to distinguish between x and y; hence, in what follows, we
use u(x, t) instead of u(y, t). Then, Eq. (24b) is written in the form
v(x, t) (x, t) = u(x, t) (x, t) for (x, t) ST .
As x F = 0 and v = u , we obtain the second kinematic condition
t F (x, t) + u x F (x, t) = 0 for F (x, t) = 0.
In the new notation, system (24c) is written in the form of the classical system of the Euler equations for the
dynamics of an ideal uid
t v + v x v + x p = 0, divx v = 0 for (x, t) QT .
Equation (24d) for u acquires the form
t u + u x u + (gtx H + (H 2 /2 + R)H) = (p + C(t)) for F (x, t) = 0.
Let us derive the transport equation for density. For this purpose, we involve several auxiliary facts of
dierential geometry. If
x(t + , ) = x(t, ) + u(x(t, ), t) + O( 2 ),
then the rst variation of the surface area is written as
d x  

g (t + , )  = tr {S(x, t)Dx u(x, t) S(x, t)} gtx ,
d =0
where S(x, t) = I (x, t) (x, t). Using the notation
divxt u = tr {S(x, t)Dx u(x, t) S(x, t)}, (25)
we obtain
t (x, t) + u(x, t) x (x, t) = t (t, )

g0  x g0
= x t g (t, ) = divxt u = divxt u,
gt gtx
which yields the transport equation for density
t (x, t) + u(x, t) x (x, t) + divxt u = 0.
As a result, we obtain the following problem equivalent to Problem A.
Problem B. We have to nd a curvilinear cylinder QT with the side boundary ST , vector elds v: QT  R3 ,
u: ST  R3 , and functions p: QT  R and : ST  R satisfying the following equations and boundary conditions:
674
t v + v x v + x p = 0, divx v = 0 for (x, t) QT ,
 (26a)
t u + u x u + (gtx H + (H 2 /2 + R)H = (p + C(t)) for F (x, t) = 0;

t + u x + divxt u = 0 for F (x, t) = 0,


(26b)
t F (x, t) + v x F (x, t) = t F (x, t) + u x F (x, t) = 0 for F (x, t) = 0.
Here the equation F (t, x) = 0 denes the surface ST ; the operator divxt is dened by Eq. (25).
These equations have to be supplemented by the initial data:
 
 
t  = , x
t = ,
t=0 t=0

v(x, 0) = v0 (x), divx v0 = 0, x ,

S(x, 0)u(x, 0) = us (x), x ,

(x, 0) = 0 (x), x .
The presence of the potential mass forces
f (x) = x (x)
does not exert any signicant eect on the form of the equations. In this case, the Lagrangians Lf and Le have the
form
 
1
Lf = |t x(t, )|2 d (x(t, )) d,
2

 
1
Le = 2
|t y(t, )| d W (yt ) (y(t, )) d.
2

Hence, in the Eulerian formulation (26), Eqs. (26a) are replaced by the equations
t v + v x v + x p + x = 0, divx v = 0 for (x, t) QT ,

t u + u x u + (gtx H + (H 2 /2 + R)H + x = (p + C(t)) for (x, t) ST .
6. Two-Dimensional Motion. In the case of two spatial variables, the equations of motion become much
simpler. With allowance for applications to the problem of surface waves in a pool covered by an elastic lm,
we assume that the domain t occupied by the uid has the form t = {x = x1i + x2j, x2 < (x1 , t)}, where
= (x1 , t) is a function periodic with respect to the variable x1 . The surface t = {x, x2 = (x1 , t)} is unknown
and is determined in the course of solving the problem.
By virtue of the assumed absence of separation of the shell from the free surface of the uid [in the plane
(i, j)], the free surface of the shell admits parametrization
t = {y: y = r(t, s), s R},
where the displacement vector r(t, s) is a periodic function of the Lagrangian variable s.
We consider auxiliary vectors a and b and new functions and :
sr b = sa ,
a = , = |sr |, = |sa |.
|sr| |sa|
Obviously, a b = a sa = b sb = 0. We can readily conclude that
sb = aa sb = a sa b = a .
In this notation, we have
  
gty (s) = (t, s),  = 1 s 1 sr = 1 sa = b,
H |H|
 =
.

675
Let t = {y = r(t, s), 0 s 2}. Let us recall that r(t, s) is a periodic function of the variable s. Without
losing generality, we assume that the following condition is satised.
Condition 1. At the initial time, the curve 0 is a straight line:
r(0, s) = si, 0 s 2.
The density 0 is already subjected to the condition of a uniform distribution of matter in the shell at the
initial time: 0 = 1. Under these conditions, the law of conservation of mass (24e) means that
1 1
(t, s) = = . (27)
|sr(t, s)| (t, s)
Let us consider the Lagrangian function Le for the membrane
Le = Ke W (t ) Ep ,
where Ke is the kinetic energy, W (t ) is the stored elastic energy, and Ep is the potential of energy due to gravity.
The kinetic energy of the elastic membrane is determined by the equality
  2
1 2 1 2 1
Ke = (t, s)|tr | dt = (t, s)|tr | (t, s) ds = |tr |2 ds.
2 2 2
t t 0

In turn, the stored elastic energy W (t ) is presented as


   2  
1 1
W (t ) = W gty (s), |H(t,
 s)| dt = W , ds.
2 2
t 0

The function E(, ) = W (, ) is subjected to the condition of convexity with respect to the variable .
For the gravity eld g = gj acting in the plane (i, j) [g = y (y ) and (y ) = gy2 ], we calculate the
potential of energy due to gravity
 2 2
Ep = (t, s)(r (t, s)) dt = (t, s)(r (t, s))(t, s) ds = (r(t, s)) ds.
t 0 0

We can easily see that the variations of the functionals Ke and Ep are determined by the equalities
2 2
Ep = g j r ds, Ke = t2r r ds.
0 0

Calculating the variation of the functional of the stored elastic energy is a more dicult problem and requires
special consideration. For brevity, the sign of the dependence on t is omitted.
We express and via new unknowns k and q:

= q, = qk.
Note that q = |sr |2 and k = |sa |2 /|sr |2 . The expression for the integral functional of elastic energy takes the
form
2
1
W (t ) = F (q, k) ds,
2
0
 
where F (q, k) = W q, k q. Then, to nd the variation

2 2 
1 1
W (t ) = F (q, k) ds = q F q + k F k ds
2 2
0 0

it is sucient to calculate q and k by the formulas


676

q = |sr |2 = 2sr s r = 2 q a s r,

|sa |2 |sa |2 1 2k
k = = 2 sr s r + 2 sa s a = sr s r
|sr |2 |sr |4 |sr |2 q
 1 
2 2k k a
+ sa s a = a s r + 2 b s s r (a, s r )
q q q q q
 1   a 
2k k k
= a s r + 2 b s s r 2 b s (a, s r )
q q q q q
 1  2k
2k k
= a s r + 2 b s s r a s r
q q q q
 1 
4k k
= a s r + 2 b s s r ,
q q q
where
1 1  sr
a = s r a a s r, sa = qk b, a = .
q q q
It follows that

1
2 2
2k  1  k 
F (k, q) ds = q q F k F a s k F b s r ds.
2 q q q
0 0

As sb = qk a, this equality can be written in the form

1
2 2
k  1  k  
F (k, q) ds = q q F k F a s k F b s r ds.
2 q q q
0 0

Let us return to the variables and in the right side of this equality. If we introduce a new function E = E(, )
as
E(, ) = F (2 , 2 /2 ) W (, /),
then the expressions at the vectors a and b are turned to
k 2 1 1
q q F k F = q F 3 k F = E(, ), q F = E(, ).
q 2 2 2
Finally, we obtain
2  
1
W (t ) = s Ua (s V )b r ds,

0

where U = E(, ) and V = E(, ).


Let the domain occupied by the uid at the initial time have the form {(1 , 2 ) R2 : 2  0}. Then,
the boundary of the domain is R. Such a choice of the domain is caused by Condition 1. We assume that

-parametrization of the domain boundary x(t, ) is chosen so that |1 x(t, 1 , 0)| = 1 for 1 R. Then, the
condition of the absence of separation of the elastic shell from the uid surface takes the form
x(t, S(t, s), 0) = r(t, s), s R,
where S(t, ): R  R is a family of dieomorphisms. Let us dierentiate this equation with respect to s:
|sr(t, s)| 1
s S(t, s) = = |sr(t, s)| = .
|1 x(t, S(t, s), 0)| (t, s)
677
In accordance with the Lagrangian variational principle, we obtain
 (s V )  
t2 r s Ua b + gj = (p(t, S(t, s), 0) + C(t))b,

where S(t, ): R  R is the dieomorphism. For p = 0 and C = 0, this equation coincides with Antmans equation
[7]. Taking into account Eq. (27) for the density , we write the last equation in the form
 t2 r  s (Ua  (s V ) b ) +  g j = (p(t, S(t, s), 0) + C(t))b.
Repeating the reasoning of Sec. 5, we present the Lagrangian function for the uid in explicit form:
 
1
Lf = |t x(t, )| d g x2 (t,  ) d.
  
2

Following the Lagrangian principle, we obtain


t2 x(t, )
 + (M 1 (t,  )) p(t,  ) + gj = 0, M = D x(t,  ).
Summarizing the results of this section, we obtain the following problem.
Problem C. We have to nd a eld of displacements of the uid x(t,  ) for  and the eld of dis-
 and C(t), and dieomorphism S(t, ): R  R to satisfy the
placements of the membrane r(t, s), functions p(t, )
equations
x(t, S(t, s), 0) = r(t, s) for s R,

b(t, s) t x(t, 1 , 0) = b(t, s) tr(t, s) for s R,
1 =S(t,s)

t2 x(t,  ) + (M 1 (t,  )) p(t,  ) + gj = 0, det M (t,  ) = 1 for  , (28)

 t2 r  s (Ua  s V b ) + gj = (p(t, S(t, s), 0) + C(t)) b for s R,

x(0,  ) =  for , r(0, s) = si, (0, s) 1 for s [0, 2],


where
1 1
M = D x(t,  ), (t, s) = = ,
|sr(t, s)| s S(t, s)

U = (W (, /)), V = (W (, /)),

sr b = sa ,
a = , = |sr|, = |sa|, = {(1 , 2 ) R2 : 2  0}.
|sr| |sa|
We write Eqs. (28) in the Eulerian coordinates. Note, as the function r(t, s) for a xed t is a 2-periodic
function, we can present the vectors a and b with the use of a new unknown function, which is the angle of
deformation (t, s):
a = cos i + sin j, b = sin i + cos j;

thereby, s = = |sa|.
We use S to denote the arc abscissa on the curve t . Then, we have
t = {y: y = x(t, S) r(t, s(S, t))}.
As |S x(t, S)| = 1, then S is a Eulerian variable. Hence, the expressions for velocity u = u(x, t), density = (x, t),
and angle of deformation = (S, t) of the elastic membrane have the form
 
 
u(x, t) = tr(t, s) ,

x=
x(t,S) s=s(S,t)
 
 
(x, t) = (t, s) ,

x=
x(t,S) s=s(S,t)
678


(S, t) = (t, s) .
s=s(S,t)

We introduce the notation




s(S, t) = a(t, s) = cos((S, t))i + sin ((S, t)) j,
s=s(S,t)


n(S, t) = b(t, s) = sin ((S, t))i + cos ((S, t)) j.
s=s(S,t)

By virtue of the kinematic condition on the free boundary, we have


 
 
 t2 r(t, s) = (t u + u x u ) .
s=const 
x=const
We can readily see that
   
   1 
(t, s)1  = (x, t) = S s(S, t), (t, s) = S (S, t) (x, t)  .
s=s(S,t) 
x=
x(t,S) s=s(S,t) 
x=
x(t,S)

Note that the parametrization s = s(S, t) was chosen with the aim of satisfying the equality
 1 

S  = s  .
S=const s=const
In what follows, we use f  instead of S f to shorten the recording. Hence, we have
1  s V  
s Ua b = (Ps Qn ) ,

where P (,  ) = 1 E(1/,  /) and Q(,  ) = 2 E(1/,  /); 1 and 2 denote dierentiation with respect to the
rst and second arguments, respectively.
Similar to derivation of the equations for Problem B, we obtain the following equations. The equation of
motion for the membrane in the Eulerian coordinates acquires the form

t u + u x u ) (Ps Qn ) + gj = (p(x, t) + C(t))n, x = x(t, S). (29a)
The function satises the law of conservation of mass
t + u x + divt u = 0 for x = x(t, S). (29b)
The pressure p and the uid velocity v = v (x, t) satisfy the equations
tv + v xv + x p + gj = 0, divx v = 0,
where t (0, T ); x belongs to a curvilinear half-plane bounded by the curve x = x(t, S). Then, the constraint
equation (absence of separation) has the form
u n = v n, x = x(t, S).
7. Steady-State Problem. Let us assume that the sought functions in the Eulerian coordinates are
independent of t. As u n = v n = 0 and x = x(S) on the free boundary, the membrane velocity can be dened as
a product of the tangential vector and an unknown scalar function u = u(S):
u x(S) = u(S)s(S).
Then, the law of conservation of mass (29b) acquires the form
(u) (S) = 0, (30a)
where (S) = x(S).
In turn, Eq. (29a) takes the form
u(us ) (Ps Qn ) + gj = (p(x(S), t) + C(t))n.
With allowance that n = s and s = n, Eq. (29a) is written in projections as
uu P  Q  + g sin = 0; (30b)

u2  P  + Q + g cos = p(x(S)) + C. (30c)


679
Equations (30a) and (30b) admit two integrals. The rst integral follows from the law of conservation of
mass
(S)u(S) = C1 = const.
Taking into account the general form of the functions P and Q,
 1    1  
P (,  ) = 1 E , , Q(,  ) = 2 E , ,

we can turn the expression P  + Q  to  , where
1  1     1    1  
(,  ) = 1 E , + 2 E , E , .

As x(S) = (x1 (S), x2 (S)) and x2 (S) = sin (S), Eq. (30b) can be written as
d 1 2 
u + (,  ) + gx2 = 0,
dS 2
whence there follows the second integral
u2 /2 = C2 (,  ) gx2 .
8. Bernoulli Law. The following expressions are valid on the free boundary x = x(S):
(S)u(S) = C1 ,
(31a)
u(S)2 = C2 2((S),  (S)) 2gx2 (S);

(S)u(S)2  (S) + (P ((S),  (S)) (S)) (Q((S),  (S))) + (S)g cos (S) = p(x(S)) + C (31b)
[S is the arc length and  (S) is the curvature]. Resolving the algebraic system (31a) with respect to the quantities
and u and substituting them into Eq. (31b), we can obtain one dynamic condition relating the boundary curvature
 to the pressure p.
This work was supported by the Russian Foundation for Basic Research (Grant No. 07-01-00309) and
Program of Integration Fundamental Research of the Siberian Division of the Russian Academy of Sciences (Grant
No. 2.1).

REFERENCES

1. M. A. Ilgamov. Introduction into Nonlinear Hydroelasticity [in Russian], Nauka, Moscow (1991).
2. J. F. Toland, Heavy hydroelastic travelling waves, Proc. Roy. Soc. London, Ser. A, 463, 23712397 (2007).
3. G. Friesecke, R. James, and S. Muller, A theorem on geometric rigidity and the derivation of nonlinear plate
theory from three dimensional elasticity, Comm. Pure Appl. Math., 35, No. 11, 14611506 (2002).
4. T. Willmore, Total Curvature in Riemannian Geometry, John Wiley and Sons, New York (1982).
5. I. Ivanova-Karatopraklieva, P. E. Markov, and I. Kh. Sabitov, Bending of surfaces. III, Fundam. Prikl. Mat.,
12, 356 (2006).
6. R. Capovilla and J. Guven, Stresses in lipid membranes, J. Phys. A, 35, 62336247 (2002).
7. S. S. Antman, Nonlinear Problems of Elasticity, Springer-Verlag, New York (1995).

680