Engineering Analysis with Boundary Elements 27 (2003) 565574 www.elsevier.

com/locate/enganabound

Boundary element simulation of large amplitude standing waves in vessels

Kenji Hamanoa,*, Sunao Murashigeb, Ken Hayamic
a

Department of Mathematical Engineering and Information Physics, Graduate School of Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan b Department of Complexity Science and Engineering, Graduate School of Frontier Sciences, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan c Mathematical Informatics Research, Foundations of Informatics Research Division, National Institute of Informatics, 2-1-2 Hitotsubashi, Chiyoda-ku, Tokyo 101-8430, Japan Received 3 September 2002; revised 4 February 2003; accepted 12 February 2003

Abstract In this paper, a uid in a vessel is considered and large amplitude standing waves (LASW) of the uid are simulated directly using the boundary element method (BEM). In the simulation, two problems come out. The rst problem is that the energy of the LASW increases gradually when using double nodes at corners in the BEM. The second problem is that projection-like proles appear near the point where the free surface meets the vessel wall when regridding is not used at each time step. These projection-like proles are not physical and indicate numerical error, and cause the simulation to break down. We found that the use of discontinuous elements solves the rst problem, and the use of the half shift technique solves the second problem. In addition, a method called RIG for highly accurate simulation using regridding is proposed and veried. q 2003 Elsevier Science Ltd. All rights reserved.
Keywords: Boundary element method; Large amplitude standing waves; Direct simulation

1. Introduction The study of large amplitude standing waves (LASW) is required in various areas. For example, melted iron containers are carried on rails in ironworks. The containers may oscillate and melted iron may spill out of the containers. Besides, for launching a rocket successfully, it is important to analyze how a liquid fuel moves inside the tank due to the rockets oscillation and how in turn that effects the rockets movement. Moreover, it is important to study the motion of liquid in ships, when the ships oscillate due to ocean waves. If the liquid in the ships starts to resonate, the ships oscillation increases rapidly and in the worst case, the ship will overturn. Ref. [12] can be cited for the direct simulation of LASW using boundary element method (BEM). They assume
* Corresponding author. Present address: Second Research Center, Technical Research and Development Institute, Japan Defense Agency, 1-2-24 Ikejiri, Setagaya-ku, Tokyo 154-8511, Japan. E-mail address: hamano@jda-trdi.go.jp (K. Hamano); sunao@sat.t.utokyo.ac.jp (S. Murashige); hayami@nii.ac.jp (K. Hayami).

periodic waves as in the formulation of Ref. [7]. They map the free surface to a closed curve in the complex plane using the assumption of periodicity. Then, they apply BEM to the closed curve in the complex plane. Most papers which simulate the dynamics of liquids in a vessel using BEM have an interest in handling sloshing analysis, and few papers simulate LASW in a vessel using BEM. Moreover, those papers which address LASW assume the periodicity in most cases, and in some cases, a vortex-sheet method is used instead of BEM [15]. On the other hand, in our paper, we directly apply BEM to a boundary of a uid in a vessel in order to simulate LASW, and do not use the periodicity. Thus, BEM has a corner problem in our simulation, and the corner points become the peak of LASW. It has been pointed out that in some applications, the intersection of the free surface and the wall (the corner point) sometimes leads to inaccurate solutions and wave instability after propagation in time [4,14]. In this respect, our simulation is more difcult than Refs. [7,12]. The overview of this paper is as follows. In Section 2, the settings of our simulation are explained. In Section 3,

0955-7997/03/$ - see front matter q 2003 Elsevier Science Ltd. All rights reserved. doi:10.1016/S0955-7997(03)00041-9

566

K. Hamano et al. / Engineering Analysis with Boundary Elements 27 (2003) 565574

potential satises the Laplaces equation, D f 0: 1 The velocity of uid particles u; vT satises the following equations: u v

f ; x f : y

2 3

u; vT is also given by the following equations

Fig. 1. Schematic illustration of the simulation.

u v

Dx ; Dt Dy ; Dt

4 5

the basic formulation is given. In Section 4, simulation results on the comparison of using double nodes and using discontinuous elements, the simulation result using the half shift technique, and the simulation result with the vertical excitation are presented. In Section 5, RIG algorithm for the highly accurate simulation is explained and veried. The simulation result with the vertical excitation using RIG is also shown. In Section 6, the large amplitude cosine wave is simulated in order to show that our method can also simulate transient standing waves.

where D=Dt is the material derivative. Positions of uid particles on the free surface x; yT satisfy the following equations: Dx f ; Dt x Dy f : Dt y 6 7

2. Settings A perfect uid contained in a rectangular vessel in two dimensional space is considered (Fig. 1). We take the x-axis in the horizontal direction and the y-axis in the vertical direction. The uid at surface in contact with air is called the free surface. The center of the vessel is x 0 m; the width of the vessel is L 2p [m], the distance between the bottom of the vessel and the still water surface y 0 m is h $ L=2: When the depth of the water is greater than or equal to half the wavelength, the motion of water beneath the surface is not affected by the bottom. (Waves under these conditions are classied as deep-water waves.) Time is denoted by t [s], the displacement of the free surface from the still water surface is h hx; t [m]. The bottom of the vessel is at y 2h [m], and the height of the vessel is 1 [m]. The gravitational acceleration is g 9.8 [m/s2]. The initial wave prole of our simulation is as shown in Fig. 1.

From Bernoullis equation, the velocity potential satises the following equation at the free surface, (    ) Df 1 f 2 f 2 2 g h: 8 2 Dt x y Using the above evolution Eqs. (6) (8), the time evolution of the free surface is simulated. f=x; f=y can be calculated using BEM as the velocity potential f satises the Laplaces equation. In this paper, the time evolution scheme based on the following Taylor expansion [9] was used: xtDt < xt Dt Dx Dt2 D2 x Dt3 D3 x Dt 2 Dt2 3! Dt3 9

Dt4 D4 x : 4! D t 4

ytDt and ftDt are expanded similarly. The second-order Lagrangian derivative is expressed as D2 x ft u u u v ; 2 x x y Dt 10

3. The numerical simulation method 3.1. The basic formulation We consider an incompressible irrotational ow. The velocity potential f can be dened. The velocity

where ft f=t:ft is also computed using BEM, since ft satises the Laplaces equation: Dft 0; and the boundary values that ft Df=Dt 2 u2 v2 on the free surface and ft =n 0 on the bottom and the wall of the vessel can be obtained. Higher order derivatives of f with respect to t are computed using BEM similarly. The reason for using the time evolution scheme based on

K. Hamano et al. / Engineering Analysis with Boundary Elements 27 (2003) 565574

567

the Taylor expansion is as follows. When the fourth-order Runge Kutta method is used for the time integration, one must compute the coefcient matrices of the system of equations in the BEM four times in each time step. On the other hand, if the Taylor expansion is used for the time integration, the computation of the coefcient matrices is required only once in each time step, so that the computation time is signicantly reduced. 3.2. BEM In this part, we briey explain the BEM [3] formulation. Let Gf be the free surface, Gw be the boundary where the uid contacts with the vessel. G Gf < Gw is a closed curve. Let V be the domain inside G: The boundary integral equation (BIE) f u f p dG f f f p dG; 2p n G G n 11

n X 1 p 2 uk fp dG fk 0 2 fk f 2 2p n Gi i1 ij ik n X i1

Gi

f p f d G; n

14

are introduced, where Gj and Gk are the two elements of G adjacent to node k: The use of discontinuous elements is as follows. The nodes for f and f=n which meet at the corner are shifted inside the two elements adjacent to the corner. f and f=n are represented by linear functions along the whole elements in terms of their nodal values and both of them are in principle discontinuous at the corner. In the present paper, when na is a node of f and f=n at the corner and nb is a node next to na ; na is shifted to the point dividing the interval na nb by the ratio 1:3.

holds, where u is the angle subtending V at the point on the boundary G; and fp is the fundamental solution. Then, G and f; f=n on G are discretized using linear elements G <i Gi : The BIE for node k is

4. Simulation results 4.1. Use of double nodes vs. discontinuous elements For initial wave prole, the large amplitude stationary standing waves obtained by Okamura [10] was used. The crest acceleration of h normalized by g Ac 21=gDfy =Dt is 0.85. Sixty-three nodes were placed on Gf ; and regridding was used at every time step. The regridding procedure is as follows. Let the x; y coordinates and the velocity potential f at node i at time t be xti ; yti ; fti i 1; 2; ; f ; respectively, where f is the number of nodes on the free surface. Using the time evolution scheme, xtiDt ; ytiDt ; fitDt i 1; 2; ; f are obtained. Then, xitDt ; ytiDt i 1; 2; ; f are interpolated using spline interpolation [11]. The y coordinate y0i of the interpolated curve at xti i 1; 2; ; f are obtained. Then, xtiDt ; ytiDt i 1; 2; ; f are substituted by xti ; y0i i 1; 2; ; f ; respectively, and ftiDt i 1; 2; ; f are substituted by   4 X 1 k t t Dt Dy fi fi i 1; 2; ; f ; 15 k! t y k1 respectively, where Dy y0i 2 yti : The point of the regridding is that the x coordinates on the free surface remain the same. The time evolution of LASW is simulated. Fig. 2 shows the evolution of the total energy. Two cases are shown. Namely, the case using double nodes and the case using discontinuous elements for corners where the free surface meets the vessel wall. It is observed that LASW simulation using discontinuous elements is more stable than using double nodes. Fig. 3 shows the magnied prole at the right end of the free surface. h hx; t is the displacement of the free surface from the still water surface. Three cases are shown.

uk f 2p k i1

n X

Gi

f dG n i1

n X

Gi

f p f d G: n

12

The BIE is then approximated by collocation. The integrations over each element are done analytically. When n nodes are placed on G; n equations are obtained. When values of either f or f=n are known at each node, unknown values of either f or f=n at each node are solved from the system of n equations because there are n unknown values. 3.2.1. The corner problem At the intersection points between Gf and Gw ; f=n has two different values. Such points are called corner nodes. At corner nodes, f=n is discontinuous and the linear interpolation of f=n causes inaccuracy. Thus, special care must be taken for the discontinuity of f=n: There are two basic methods for this in BEM. The rst method is to use double nodes [6] and the second method is to use discontinuous elements [2]. The use of double nodes is as follows. The corner node where Gf and Gw meet is split into two nodes k; k0 ; and the values of f and f=n at the two nodes are treated as f f 0 independent quantities fk ; fk0 ; n lk and n lk : Because two f f of fk ; fk0 ; n lk and n lk0 are known values and two of f f 0 fk ; fk0 ; n lk and n lk are unknown values, another two equations

fk fk0 ;

13

568

K. Hamano et al. / Engineering Analysis with Boundary Elements 27 (2003) 565574

Fig. 2. Evolution of the total energy of the standing waves. Two cases are shown. Namely, the case using double nodes and regridding (double node, regrid) and the case using discontinuous elements and regridding (discontinuous, regrid).

Fig. 4. Three wave proles when the simulation without using regridding fails. Time progresses as the index number increases.

Namely, the case using double nodes, the case using discontinuous elements, and the case using the method of Ref. [7] (250 nodes were placed on Gf in this case). Note that when using the method of Ref. [7], there is no corner problem since it assumes periodicity. The effect of the corner nodes can be observed by comparing the three cases. It is observed that there is a little declination between the prole using double nodes and using the method of Ref. [7], whereas, the prole using discontinuous elements matches well with that using the method of Ref. [7]. As a result, we recommend that discontinuous elements should be used when simulating LASW with BEM. 4.2. The half shift technique Next, we consider the use of methods which can simulate LASW without using regridding. The reason considering this is that when the amplitude of the standing waves becomes large, the accuracy of spline interpolation used in

regridding may deteriorate, so that, if we want to simulate the limit amplitude standing waves, the simulation without using regridding may be required. When the initial wave prole was set as Fig. 1, that is, when the center displacement of the free surface from the still water surface at t 0 : h0; 0 is at the highest and hp; 0; h2p; 0 are at the lowest, and LASW is simulated without using regridding, projection-like proles appear at the end of the free surface as shown in Figs. 4 and 5. These projection-like proles are not physical and indicate numerical error, and that causes the simulation to break down. Then, we consider two initial wave proles. For the rst prole, h0; 0 is at the highest and hp; 0; h2p; 0 are at the lowest (no_shift), for the second, h0; 0 is at the lowest and hp; 0; h2p; 0 are at the highest (half_shift) (Fig. 6). In both cases, 63 nodes were placed on the free surface. When the initial wave prole is set to no_shift of Fig. 6, the evolution of the free surface is as shown in Fig. 7. This simulation fails before half the period of oscillation, and nodes are found to concentrate towards the end of the free surface.

Fig. 3. Wave proles for three cases: using double nodes and regridding (double node, regrid), using discontinuous elements and regridding (discontinuous, regrid) and using the method of Ref. [7] and without using regridding with 250 nodes on the free surface (L-H, C, noregrid, 250points).

Fig. 5. Magnication of Fig. 4.

K. Hamano et al. / Engineering Analysis with Boundary Elements 27 (2003) 565574

569

Fig. 6. The initial amplitude of the free surface hx; 0 Ac 0:85:

Fig. 8. Evolution of the wave prole. The initial amplitude of the free surface is set to half_shift of Fig. 6.

When the initial wave prole is set to half_shift of Fig. 6, the evolution of the free surface is as shown in Fig. 8. Different from the simulation of no_shift, this simulation does not fail, and the nodes do not concentrate towards the end of the free surface. Therefore, shifting the initial wave prole by half the wave length prevents the concentration of nodes at the end of the free surface. This, in turn, enables us to avoid the formation of projection-like proles, and renders the simulation without regridding feasible. 4.3. Simulation of excited standing waves In this part, we excite the LASW with Ac 21=g Dfy =Dt 0:85 in the vertical direction and make the LASWs amplitude larger and larger. The initial wave prole was set to the half_shift given in Fig. 6, and discontinuous elements were used. The oscillating amplitude was d 0:01 m: Since the depth of the vessel h is sufciently large, the angular frequency v of p the free surface obtained from linear analysis is g 1=s: According to Mathieus theory [1,8,15], when the angular frequency of the forced vertical oscillation is set to twice of v; the amplitude of the standing waves becomes

larger and larger. The oscillation is represented as follows: gt g d2v2 sin2vt m=s2 : 16

Fig. 9 shows the evolution of Ac ; of the acceleration of h0; t normalized by g; using regridding. Two cases are shown. Namely, the case when 127 and 251 nodes were placed on the free surface, respectively. The results for both cases are clearly different and there are some disturbances. Fig. 10 shows the evolution of Ac without using regridding. Two cases are shown. Namely, the case when 127 and 251 nodes were placed on the free surface, respectively. The results for both cases are close, and the disturbances seen in Fig. 9 have diminished. As a result, in order to simulate the limit LASW, we recommend that regridding should not be used. Spline interpolation in the regridding procedure might be causing the disturbances.

Fig. 7. Evolution of the wave prole. The initial amplitude of the free surface is set to no_shift of Fig. 6.

Fig. 9. Evolution of the amplitude of the crest acceleration normalized by the gravitational acceleration Ac : The case with 127 nodes and 251 nodes on the free surface are shown, respectively. In each case, regridding was used.

570

K. Hamano et al. / Engineering Analysis with Boundary Elements 27 (2003) 565574

the expansion, determine Ak k 0; 1; ; kmax21 from xi ; yi ; fi ; i 1; 2; ; f using the least square method. The appropriate value for kmax requires trial and error. Here, we set it to bf 1=2c: The expansion (17) is derived from the Laplaces equation: Df 0 and the bottom and wall boundary conditions: f= n 0: 4. Execute the convergence computation with respect to yi i 1; 2; ; f using the Newton method and Bernoullis equation as follows. for i 1 to f while yi converges do
Fig. 10. Evolution of the amplitude of the crest acceleration normalized by gravitational acceleration Ac : The case with 127 and 251 nodes on the free surface are shown, respectively. In each case, regridding was not used.

yi yi 2

2 2 ft 1 2 fx fy gy : fty fx fxy 2 fy fxx g

5. The RIG method If we want to simulate not only standing waves but more general waves like transient waves, regridding may be required. But regridding with simple spline interpolation gives rise to numerical error. In this part, we propose a method called RIG in order to make the wave simulation more accurate. The idea is based on Refs. [10,13]. The basic idea is as follows. First, the velocity potential is expanded using the basic functions which satisfy the Laplaces equation. Then, these coefcients are corrected to satisfy Bernoullis equation rigorously. Note that this proposed method can solve the problem which has symmetrical geometry and uniform depth. 5.1. The RIG algorithm The position of the node i on the free surface is xi ; yi T ; and the velocity potential of the node i is fi : The number of nodes on the free surface is f : The following algorithm gives the time evolution from t to t Dt: Assume that at time t; xi ; yi ; fi i 1; 2; ; f satisfy Bernoullis equation (8) rigorously. 1. Using a time evolution scheme such as in Section 3.1, compute xi ; yi ; fi i 1; 2; ; f at t Dt: 2. Regrid the free surface so that xi remains the same for each node i i 1; 2; ; f : Update yi and fi i 1; 2; ; f using interpolation. 3. Using the expansion of the velocity potential [5,13]

In order to calculate ft ; the time derivative of Ak k 0; 1; ; kmax21 are required. To calculate the time derivative of Ak ; we must store the Ak k 0; 1; ; kmax21 of the previous time steps, and approximate as follows,
t 2 2 Dt t23Dt dA k 11Atk 2 18Atk2Dt 9Ak 2 2 Ak < : dt 6D t

18

5. When yi 1; 2; ; f 6. The free derivative

i 1; 2; ; f converge, compute fi i using Eq. (17). surface should also satisfy the material of Bernoullis equation, 19

ftt 2ftx fx 2fty fy 2fxy fx fy

2 fxx f2 x 2 fy gfy 0:

In order to compute Eq. (19), d2 Ak =dt2 is approximated by

t 2 3 Dt d 2 Ak 2Atk 2 5Atk2Dt 4Atk22Dt 2 Ak < : dt 2 Dt2

20

The left-hand side of Eq. (19) is written as DPDT xi ; yi ; {Ak } for brevity. This is a function of xi ; yi and {Ak } {A0 ; A1 ; A2 ; ; Akmax21 }: Then, {Ak } are determined using Newtons method, so that, for each node i 1; 2; ; kmax ; DPDT is sufciently close to 0. To do so, the Jacobian matrix J Jij i; j 1; 2; ; kmax is required, where DPDTxi ; yi ; {Ak } : 21 Jij Aj21 Since it is complicating to compute the Jacobian matrix analytically, Jij is approximated by Jij DPDTxi ; yi ; {Ak } 1 2 DPDTxi ; yi ; {Ak } ; 1 22

fx; y

kmax 21 X k0

cosh ky h coskx; Ak coshkh

17

where kmax is the truncation number of the summation and Ak k 0; 1; ; kmax21 are the coefcients of

where {Ak } 1 {A0 ; ; Aj22 ; Aj21 1; Aj ; ; Akmax21 } and 1 1026 : Then, execute Newtons method as follows.

K. Hamano et al. / Engineering Analysis with Boundary Elements 27 (2003) 565574

571

until DPDT converges to 0 at every node i 1 0 1 1 0 A0 DPDTx1 ; y1 ; {Ak } A0 C B C C B B B A C B A C B DPDTx ; y ; {A } C C B 1 C B 1 C B 2 2 k C B C C B B C: B . C B . C 2 J 21 B . C B . C B . C B . C B . C B . C B . A @ A A @ @ Akmax21 Akmax21 DPDTxkmax ; ykmax ; {Ak } 0 When the free surface is symmetric with respect to the y-axis and kmax bf 1=2c; the above Newtons method can be executed. 7. If the convergence is sufcient in the steps 4 and 6, go to the step 8, otherwise go to step 4. 8. t t Dt: 5.2. Verication of the RIG method Here, we compare results using RIG with results obtained without RIG. The initial wave prole is given by hx; 0 0:1 cos x with Dt xed. The time evolution scheme based on Taylor expansion was used. Fig. 11 shows the uctuation of the total volume. Four cases are shown. Namely, the case when using RIG with Dt 0:01 s RIG; dt 0:01; when using RIG with Dt 0:001 s RIG; dt 0:001; when not using RIG with Dt 0:01 s noRIG; dt 0:01; and when not using RIG with Dt 0:001 s noRIG; dt 0:001: Nineteen nodes were placed on the free surface and 9 nodes on the left and the right wall of the vessel, respectively, and 19 nodes on the bottom line of the vessel. The depth of the vessel is h p [m]. It is observed in this gure that when using RIG, the uctuation of volume is smaller than when RIG was not used. Similarly, Fig. 12 shows the uctuation of energy. It is observed that when using RIG, the uctuation is larger than

Fig. 12. Evolution of the uctuation of energy when the initial amplitude of the free surface is given by hx; 0 0:1 cos x: Four cases are shown: using RIG with Dt 0.01 [s] RIG; dt 0:01; using RIG with Dt 0.001 [s] RIG; dt 0:001; not using RIG with Dt 0.01 [s] noRIG; dt 0:01; and not using RIG with Dt 0.001 [s] noRIG; dt 0:001: Nineteen nodes were placed on the free surface.

when RIG was not used. However, when using RIG, the uctuation has neither an increasing trend nor a decreasing trend. 5.3. Simulation of excited standing waves using RIG Next, using the RIG method, we will simulate the behavior of standing waves when they are excited in the vertical direction so that the amplitude of the waves becomes larger and larger. The initial wave prole is set to hx; 0 0:1 cos x: The waves are excited intermittently with a period of four seconds. Thus, the commencement of each excitation t0 is 4j j 0; 1; 2; : For an example, we excite during 0 # t , K ; 4 # t , 4 K ; 8 # t , 8 K ; and we do not excite during K # t , 4; 4 K # t , 8; 8 K # t , 12; :

Fig. 11. Evolution of the uctuation of the volume when the initial amplitude of the free surface is given by hx; 0 0:1 cos x: Four cases are shown: using RIG with Dt 0.01 [s] (RIG, dt 0:01), using RIG with Dt 0.001 [s] (RIG, dt 0:001), not using RIG with Dt 0.01 [s] (noRIG, dt 0:01), and not using RIG with Dt 0.001 [s] (noRIG, dt 0:001). Nineteen nodes were placed on the free surface.

Fig. 13. Evolution of h0; t for the case when the initial amplitude of the free surface is set to hx; 0 0:1 cos x and RIG is used. The rectangular vessel is excited intermittently in the vertical direction.

572

K. Hamano et al. / Engineering Analysis with Boundary Elements 27 (2003) 565574

Fig. 14. Time evolution of the free surface from top to bottom, and then left to right. The initial amplitude of the free surface is given by hx; 0 20:5 cos x:

The gravitational acceleration, which is a function of time gt; is given by 8 2 2 > < g d 2v2 T T 2 K sin2vT ; 0 # T , K ; g t 16 > : g; K # T , 4; 23 where g 9.8 [m/s2], ldl 0.01 [m]. The sign of d is determined so as not to decrease the energy of the wave. v is the angular frequency of the free surface, and

K 2p=v 2 s: T t 2 t0 is the time from the most recent commencement of excitation. T 2 T 2 K 2 =16 is introduced for the smooth shifting from the excitation period to the non-excitation period and vice versa. Sixty-three nodes were placed on the free surface and the time evolution scheme based on Taylor expansion was used. Dt was xed at 0.01 [s]. RIG was used. Fig. 13 shows the evolution of h0; t: It is observed that the simulation fails before max{h} $ 0:3 m: In the RIG algorithm, higher order approximations 24

t 2 2 Dt t23Dt t24Dt t 2 5 Dt dA k 137Atk 2 300Atk2Dt 300Ak 2 200Ak 75Ak 2 12Ak < ; dt 60Dt

t 2 3 Dt t25Dt d 2 Ak 45Atk 2 154Atk2Dt 214Atk22Dt 2 156Ak 61Atk24Dt 2 10Ak < ; dt 2 12Dt2

25

K. Hamano et al. / Engineering Analysis with Boundary Elements 27 (2003) 565574

573

Fig. 15. Time evolution of the free surface from top to bottom, and then left to right. The initial amplitude of the free surface is given by hx; 0 20:5 cos x:

were used instead of Eqs. (18) and (20) in order to obtain better results. The reason why the simulation fails might be due to the inaccuracy of the approximation of time derivatives of the velocity potential ft ; fty ; etc.

6. Simulation of large amplitude cosine wave Using the simulation method of this paper, transient standing waves can be simulated. For example, the large amplitude cosine wave is simulated in this section. The initial wave prole is given by hx; 0 20:5 cos x: No excitation is applied. Discontinuous elements were used at the wall free surface interface in BEM. The time evolution scheme based on Taylor expansion was used. RIG

was not used. Fifty-nine nodes were placed on the free surface, and regridding was not used. Fig. 14 shows the wave prole at t 0, 0.0659, 0.352, 0.414, 0.486, 0.589 [s], from top to bottom, and then left to right, respectively. Fig. 15 shows the wave prole at t 0.786, 1.07, 1.38, 1.65, 1.92, 2.10 [s], from top to bottom, and then left to right, respectively. The last gure of Fig. 15 shows that the free surface has a very sharp prole. Therefore, we cannot simulate beyond this point. In this case, regridding along the free surface may be necessary.

7. Conclusions In this paper, LASW in a vessel were simulated directly using the BEM. In the simulation, two problems occurred. The rst problem was that when double nodes were used at

574

K. Hamano et al. / Engineering Analysis with Boundary Elements 27 (2003) 565574

corner nodes where the free surface meets the walls of the vessel, the total energy tended to increase gradually. The second problem was that when not using regridding, projection-like proles appeared at the end of the free surface which caused the simulation to break down. The following methods were proposed to overcome these problems. By using discontinuous boundary elements at the corner nodes, the gradual increase of the energy was avoided. By setting the initial wave prole so that the lowest point was at the middle and the highest point at the ends of the free surface, the nodes on the free surface did not concentrate at the ends of the free surface, whereas when the initial wave prole has the lowest point at the ends of the free surface and the highest point at the middle of the free surface, the nodes on the free surface concentrated at the ends of the free surface causing projection-like proles and the simulation to break down. Hence, the treatment of corner nodes where the free surface meets the vessel is very signicant for the stable simulation of LASW. Simulating without regridding is important when simulating limit amplitude standing waves. However, for situations where regridding is required, we proposed the simulation method RIG, for simulating the free surface with high accuracy. However, RIG still has some problems when simulating LASW. In order to investigate this problem of RIG, validated computation using interval arithmetic may be required. Finally, the large amplitude cosine wave was simulated without regridding.

We thank Dr T. Nakayama of Chuo University, for his detailed explanation of the time evolution scheme based on Taylor expansion. We also thank Dr K. Abe of Niigata University, for his enlightening advice and discussions.

References
[1] Benjamin TB, Ursell F. The stability of the plane free surface of a liquid in vertical periodic motion. Proc R Soc Lond 1954;A255: 505 15. [2] Brebbia CA, Dominguez J. Boundary elements: an introductory course. Boston: Computational Mechanics Publications; 1992. [3] Enokizono M. Boundary element analysis (in Japanese). Tokyo: Baifukan; 1986. [4] Grilli ST, Svendsen IA. Corner problems and global accuracy in the boundary element solution of nonlinear wave ows. Engng Anal Bound Elem 1990;7(4):178 95. [5] Hammack JL. A note on tsunamis: their generation and propagation in an ocean of uniform depth. J Fluid Mech 1973;60(4):769 99. [6] Hartmann F. Introduction to boundary elements: theory and applications. Berlin: Springer; 1989. [7] Longuet-Higgins MS, Cokelet ED. The deformation of steep surface waves on water: I. A numerical method of computation. Proc R Soc Lond 1976;A350:126. [8] McLachlan NW. Theory and application of Mathieu functions. London: Oxford University; 1947. [9] Nakayama T. A computational method for simulating transient motions of an incompressible inviscid uid with a free surface. Int J Numer Meth Fluids 1990;10:68395. [10] Okamura M. The prole of large amplitude standing waves (in Japanese). RIMS Kokyuroku 1997;993:134 43. [11] Press WH, Flannery BP, Teukolsky SA, Vetterling WT. Numerical recipes in C. Cambridge: Cambridge University Press; 1993. [12] Saffman PG, Yuen HC. A note on numerical computations of large amplitude standing waves. J Fluid Mech 1979;95(4):70715. [13] Tsai CP, Jeng DS. Numerical Fourier solutions of standing waves in nite water depth. Appl Ocean Res 1994;16:18593. [14] Tsai W-T, Yue DKP. Computation of nonlinear free-surface ows. Annu Rev Fluid Mech 1996;28:24978. [15] Wright J, Yon S, Pozrikidis C. Numerical studies of two-dimensional Faraday oscillations of inviscid uids. J Fluid Mech 2000;402:132.

Acknowledgements We are very grateful to Dr M. Okamura of Kyushu University, for his large amplitude stationary standing wave proles data which were needed for the initial wave proles in this paper.