This action might not be possible to undo. Are you sure you want to continue?

Mark A. Christony Computational Physics R&D Department Sandia National Laboratories M/S 0819, P.O. Box 5800 Albuquerque, New Mexico 87185-0819 PH.: (505) 844-8520 FAX: (505) 844-0918 E-mail: machris@sandia.gov submitted to: Computer Methods in Applied Mechanics and Engineering Submitted: September 5, 1995 Revised: December 22, 1997

y

This work was supported by the U.S. D.O.E. under Contract DE-AC04-94AL85000. Previously at Lawrence Livermore National Laboratory.

The application of discrete solution methods to the second-order wave equation can yield a dispersive representation of the non-dispersive wave propagation problem resulting in a phase speed that depends not only upon the wavelength of the signal being propagated but also upon the direction of propagation. In this work, the dependence of the dispersive errors on the wave propagation direction, mesh aspect ratio, and wave number is investigated with the goal of understanding and hopefully reducing the phase and group errors associated with the two-dimensional bilinear nite element. An analysis of the dispersive e ects associated with the consistent, row-sum lumped and higher-order mass matrices has led to a reducedcoupling \penta-diagonal" mass matrix that yields improved phase and group errors with respect to wavelength and propagation direction. The in uence of row-sum lumping the nite element mass matrix is demonstrated to always introduce lagging phase and group error, while a linear combination of the lumped and consistent mass matrices, i.e., a higher-order mass matrix, is shown to improve the dispersion characteristics of both the reduced and full-integration element. Using a 2 5% phase error metric, the higher-order mass matrix can reduce the mesh resolution requirements relative to the lumped mass matrix by nearly a factor of two in onedimension; a factor of 4 in two-dimensions. Similarly, the reduced-coupling mass matrix can reduce the grid requirements by a factor of approximately four relative to the lumped mass matrix; a factor of 16 in two-dimensions. Although the reduced-coupling mass matrix requires additional storage relative to the lumped mass, the factor of 16 reduction in grid requirements promises tremendous overall savings for computing a xed wavelength spectrum, or alternatively for resolving shorter wavelengths with a xed grid resolution.

:

Abstract

1 Introduction

The numerical solution of the second-order, wave equation arises in many physical problems of engineering interest. One such application is acoustic uid-structure interaction where an acoustic uid surrounds and interacts with an elastic body 1]. For this class of problems, complex geometry can have a signi cant e ect upon the scattered-radiated pressure eld. The nite element method is the natural choice for these problems because of its ability to treat arbitrarily complex geometry. However, the nite element method, like most discrete solution methods, is limited in its ability to resolve a broad spectrum of signals with a xed grid resolution due, in part, to dispersive errors. In general, the application of discrete methods to solve the wave equation results in solutions that are dispersive even though the physical model for wave propagation is nondispersive. In order to apply grid-based numerical methods such as the nite element method (FEM) or the nite di erence method (FDM) to the wave equation and faithfully represent the physics, a clear understanding of the phase and group errors associated with the numerical method is necessary. At a minimum, knowledge of the dispersive characteristics for a discrete solution method enables the construction of meshes that can accurately resolve the wavelength spectrum of interest. This knowledge enables the

analyst to avoid introducing signals that cannot be represented on the discrete physical domain. Understanding and extending the spectrum of signals that can be resolved with the nite element method for a xed grid resolution is one of the goals for this work. Dispersion error is characterized by the di erence between the apparent, i.e., numerical, phase and group speed of waves and their exact counterparts. In the context of linear acoustics, the phase speed is the speed at which individual waves propagate. In the absence of dispersion, i.e., for a perfect acoustic uid, this is simply the sound speed. In a dispersive medium, the phase speed is a function of the frequency or wavelength of the propagating wave. In this context, phase error may be viewed as a measure of the in uence of numerical dispersion on the apparent sound speed relative to the true sound speed. In contrast to the phase speed, the group speed describes the propagation of wave packets that are comprised of short wavelength signals modulating a longer wavelength envelope. Thus, the group speed refers to the speed of a wave group, and is sometimes referred to as the speed of modulation. The energy associated with a wave packet travels with the packet and often the group speed is referred to as the \energy" velocity. For a non-dispersive acoustic medium the phase and group speed are identical. However, in discrete solutions, the group speed may be used to study and explain the propagation of the short wavelength, spurious oscillations that are typically 2 x in wavelength. Here, x is the mesh spacing in the x-coordinate direction. Vichnevetsky 2] has demonstrated that the spurious 2 x oscillations, that can be induced by rapid changes in mesh resolution or physical boundaries, propagate at a group speed associated with a 2 x wavelength. This paper focuses upon the dispersive errors introduced when using a nite element spatial discretization for the second-order wave equation by examining the phase and group error associated with the bilinear quadrilateral element. This e ort is an extension of the work by Mullen and Belytschko 3] for the quadrilateral element because it characterizes both the phase and group errors with respect to lumped, consistent, higher-order, and reduced-coupling mass matrices for both full and reduced integration. Although this work does not consider the fully discrete system as in Schreyer 4], it does consider both the phase and group speed for a range of grid aspect ratios in two-dimensions and proposes a modi ed mass matrix to improve the dispersive characteristics of the bilinear element. Before proceeding with a discussion of the numerical analysis, a brief historical review of investigations into the dispersive errors associated with discrete solution methods is presented. The e ects of consistent, lumped and higher-order mass matrices on the phase speed for linear and quadratic nite elements were investigated by Belytschko and Mullen 5] for linear elastic wave propagation in one dimension. Here, the compensatory interaction between the time integrator and mass matrix, i.e., trapezoidal rule and consistent mass versus central di erence and lumped mass, was veri ed, but shown to depend upon the Courant number. Vichnevetsky 6, 7, 8] has investigated the dispersive nature of both nite di erence and nite element methods for the rst-order wave equation. In 9], the dispersive errors introduced by nonuniform grid spacing and \hard" boundaries are discussed, and the

higher-order. Shakib and Hughes 16] have applied Fourier analysis to the space-time Galerkin least-squares method for advection-di usion problems. the two-dimensional wave equation is ( ) @ 2 p = c2 @ 2 p + @ 2 p . More recently. Trefethen 11] has considered the role of group velocity in understanding the propagation of wave packets. the results of the dispersion analysis are summarized and conclusions are drawn. and reduced-coupling mass matrices are presented. (1) @t2 @x2 @y2 where p is the perturbation pressure. For this application. in part. i.possibility of using arti cial viscosity to damp short wavelength spurious waves is investigated. This analysis technique was applied by Park and Flaggs 14] in an e ort to understand and ameliorate locking phenomena in C o plate elements. the generation of parasitic waves at interfaces. Thompson and Pinsky 19] extended the concepts of Fourier analysis in order to treat p-version nite element discretizations. In section 3. time-domain. lumped.e. The phase and group speed for both full and reduced integration elements with the consistent. PING 1]. . and c is the uid sonic velocity. 2 Governing Equations and Analysis The motivation for this investigation is based. Finally. and stability. In this context. Similarly. This work provides practical guidelines for the number of elements per wavelength in terms of the spectral order. Grosh and Pinsky 20] have applied Fourier dispersion analysis to uid loaded plates for structural acoustics simulations.. Karni 12] has characterized the group speed errors associated with symmetric upwind schemes for pure advection. upon the structural acoustics code. Fourier analysis has also been applied to nite element discretizations in order to understand the dispersive nature of elastic wave propagation in bars and locking phenomena in beams 13]. Harari 17] presents the phase error associated with the Galerkin least-squares discretization for the second-order wave equation in a nite domain. Deville and Mund 18] have used Fourier analysis to investigate the spectral behavior of the iteration matrix for nite element preconditioning. a rst order wave equation. acoustic uid-structure interaction problems. The following discussion begins with an overview of the dispersion analysis and a summary of the formulae for computing the phase and group speed for FDM and FEM semi-discretizations. the dispersion characteristics for the nite di erence method are presented rst as a basis for comparison with the nite element discretization. the perturbation pressure is very natural to work with because it may be used to directly compute the uid loads on the structure. Alvin and Park 15] have also used Fourier analysis to tailor the frequency response of beams and bars discretized with the nite element method. which uses the nite element method for the solution of fully-coupled. the in uence of group velocity in two-dimensional nite di erence discretizations with uniform aspect ratio was considered. Similar analysis techniques have been applied to wave propagation in periodic domains 10]. Here.

and are not repeated here. the reduced-integration \acoustic" element in PING 1. higher-order and reduced-coupling mass matrices are considered. Following Mullen and Belytschko 3]. (2) where M and K are the generalized mass. For each nite element formulation. reduced-integration is taken to be synonymous with the use of single-point Gaussian quadrature for the calculation of the element mass and sti ness operators. (3) where Mc is the consistent mass. the generalized mass matrix is a linear combination of the consistent and lumped mass matrices. The wave vector is ~ = kcos( ) ^ + ksin( ) ^. and the nite element method with both full and reduced-integration elements. In the case of the nite di erence discretization. M is the identity matrix.e. with the aspect ratio. nite di erence operator. In the Fourier analysis. k denotes the magnitude of the wave vector. The semi-discrete form of equation (1) is 2.. ! is the wave frequency. = y= x. a plane wave solution to equation (1) is used in the following form. For the nite element method. and y = n y. This is illustrated in Figure 1a for a patch of four quadrilateral elements and in Figure 1b for a ve-point nite di erence stencil. i. The Fourier analysis used to evaluate the phase and group speed is summarized here with the dispersion formulae for the nite di erence method. 106 elements can be advanced one time step in a single cpu second. p (4) .n = Aexp ik(m xcos + n ysin ) ? i!t]: Here. and K is the penta-diagonal.1 Fourier Analysis M p + Kp = F. and sti ness respectively.Wallace 21] presents alternative acoustic formulations that also result in a secondorder wave equation. 22] requires approximately 1 s of cpu time on a CRAY Y-MP to advance one element one time step. Although motivated by the acoustic uid-structure problem. lumped. albeit with a di erent dependent variable. In this work. where ^ and ^ are unit vectors in the x and k i j i j y-coordinate directions respectively. Pm. For example. a regular grid is considered where the mesh spacing in the x and y-coordinate directions is x and y respectively. and the nodal x and y locations are given by x = m x. The details for obtaining the weak form of the wave equation and the associated mass and sti ness operators are well known 23]. the consistent. Ml is the row-sum lumped mass matrix. The wave propagation direction is denoted by . M = Mc + (1 ? )Ml . and 0 1. Attention is given to the reducedintegration formulation because of the great computational e ciencies that can be derived from careful implementation of this element. the following analysis and results are appropriate for any physical problem governed by the second-order wave equation. second-order. and i = ?1.

The group velocity is not aligned with the wave vector. vg = vg x cos( ) + vg y sin( ). aspect ratio. (8) dk in order to simplify the requisite computations. In two-dimensions. In the discussion that follows. 6= since the phase velocity is aligned with the wave vector. and vg y = @!=@ky . the group velocity is vg = @k ^ + @k ^: ~ @! i @! j (6) x y Here. e. the group speed in the wave vector direction is de ned as vg = d! ..In the ensuing discussion. but the group velocity is oriented in the direction normal to the circular frequency. but also that the propagation direction for the group velocity depends upon the grid orientation. \group speed" will refer to the magnitude of the group velocity in the direction. Here. In general. the following relations are de ned. the work of Mullen and Belytschko 3]. The group velocity in the direction of the wave vector is simply. and is presented simply as vg . . !. where vg x = @!=@kx . the phase speed is de ned as vp = ! : (5) k The phase velocity is oriented in the direction of the wave vector. the projection of the group velocity onto the direction of the wave vector is used as a simpli ed metric for group errors. and vp represents the magnitude of the phase velocity vector. Thus. but instead has a propagation direction de ned by vg y ! = arctan (7) vg x . and wave number. In order to simplify the subsequent analysis and presentation.g. For the sake of comparison to previous work. the component of the group velocity that is orthogonal to the phase velocity is neglected. This suggests that not only are the phase and group velocity functions of wave number and orientation with respect to the grid. kx = kcos( ) and ky = ksin( ).

n) is ( x2 16P + 4(P m. are 2( 2 + 1).n?1 + Pm?1.n?1 + Pm+1. analogous to equation (11). .n m.n+1 + Pm+1.n?1 + Pm.n+1 + Pm+1. n). while the non-zero entries in the sti ness matrix. ? 2 and ?1. the lumped mass and sti ness matrices. M . is simply the identity. The full-integration elementlevel operators may be found in equations (24)-(27) in Appendix A. Turning to the nite element discretization.n+1 ) 36 +(( m?1.n + Pm.n?1 + Pm+1.n + Pm+1. Thus.f1 f2 f3 f4 g1 g2 g3 g4 = = = = = = = = cosfk cosfk cos(k cos(k x(cos + sin )g x(cos ? sin )g x sin ) xcos ) (9) ? ? ? ? x(cos + sin )sinfk x(cos + sin )g x(cos ? sin )sinfk x(cos ? sin )g x (sin )sin(k x sin ) x(cos )sin(k xcos ) (10) Semi-Discrete Equations To begin the Fourier analysis.n?1 + Pm?1. the requisite element level operators are the consistent mass.n?1 + Pm. the FEM assembly procedure 23] uses local nodes n1 through n4 in each of the four elements shown in Figure 2.n+1)] = 0: ) (12) The in uence of the consistent mass matrix is apparent in the broad 9-node coupling of the second-order time derivatives in equation (12). K . the mass matrix. The row-sum lumping procedure yields a diagonal mass matrix for the full-integration element with global entries of x2 .n + Pm+1.n + Pm+1.n ? (1 + 2)(Pm?1.n ) ? (Pm. the complete semi-discrete nite di erence equation at node (m. n) is ) ( c 2 2( 2 + 1)P ? 2(P Pm.n+1)] P ) 2 + c 8(1 + 2)Pm. the complete semi-discrete nite element equation at node (m. To obtain a complete equation at node (m.n+1) 6 +2(?2 2 + 1)(Pm?1.n m?1.n+1 )] = 0: (11) x Here.n + m.n) + 2( 2 ? 2)(Pm. after assembly of the consistent mass and sti ness.n?1 + Pm?1.

full and reduced-integration nite element formulations. substituting the plane . n 1) grid locations. does not improve the dependence on propagation direction. The consistent. This suggests that the nine-point coupling of the consistent mass matrix contributes to the angular sensitivity.n ? (1 + 2)(Pm?1.n+1) 6 +2( 2 ? 2)(Pm. lumped (LM). nine-point spatial coupling of time-derivative terms introduced by the consistent mass matrix produces leading phase and group error when the wave propagation direction is aligned with the grid.n)] = 0: (13) ) The same procedure may be used to generate complete semi-discrete equations at node (m.n + Pm+1. the complete. The detailed investigation of the phase and group error for the quadrilateral element led to the observation that the broad. semi-discrete nite element equation at node (m. a partial rowsum lumping is used to construct a penta-diagonal mass matrix. shown in Figure 1a. a linear combination of the lumped and consistent mass matrices. lumped mass (LM) and generalized mass (GM) are presented in Table (1). lumped.n + Pm+1. After the nite element assembly of the fully-integrated sti ness and reduced-coupling mass. n) is ( x2 20P + 4(P m.n+1 + Pm+1. attention may be turned to computing the dispersion formulae for each of the numerical methods. generalized (GM) and reduced-coupling (RC) mass matrices.n+1 )] 36 ( ) m.n + Pm. The entries for the global sti ness operators are summarized in Table (2) for the nite di erence. Returning rst to the computation of the phase and group speed for the nite di erence method. Appendix A.n?1 + Pm+1.e. i. n) using the reduced-integration quadrilateral element. to the diagonal of the consistent mass matrix. the higherorder mass.n 2 + c 8(1 + 2)Pm.n?1 + Pm?1. Although this suggests that the partial row-sum lumping operation must occur in the global mass matrix. n) in the global mass matrix are summarized in Table (1) for the consistent (CM). and reduced-coupling mass and sti ness operators for the reduced-integration element are presented in Appendix A.n?1 + Pm.n+1) + 2(?2 2 + 1)(Pm?1. In order to test the hypothesis that the nine-point spatial coupling of the consistent mass matrix increases the angular sensitivity of the dispersion errors. the use of a lumped mass matrix results in lagging phase and group error for all propagation directions. the partial lumping is achieved using the reduced-coupling element mass matrix in equation(26). The partial row-sum lumping simply adds the global mass matrix contributions from the (m 1. Further.The global mass matrix entries corresponding to node (m. n) for the consistent mass (CM).n?1 + Pm?1. Dispersion Formulae With the semi-discrete operators de ned. The resulting entries for node (m.. In contrast.

(8) and (14). the non-dimensional group speed.n Mm. is q (14) = k 1 x 2 (1 + 2) ? f3 ? 2f4 ]: Using equations (5). and making use of equation (5).n Km 1. and RC the reducedcoupling mass matrix. CM denotes the consistent mass.n 1 1 0 0 x2 =36 16 4 4 2 x 1 0 0 2 =36 16 + 36(1 ? ) x 4 4 x2 =36 20 4 4 2 =16 x 4 2 2 2 x 1 0 0 2 =16 4 + 16(1 ? ) x 2 2 x2 =16 8 2 2 1 Mm 0 1 0 0 1 0 0 1.n Mm 1.n 1 Km.n 2 ?2( 2 + 1) 2 Finite Di erence fc= xg 0 1 2 =6 2 ) 2(?2 2 + 1) 2( 2 ? 2) ?(1 + Full-Integration c 8(1 + 2 =4 Reduced-Integration c 4(1 + 2 ) 2(? 2 + 1) 2( 2 ? 2) ?(1 + Table 2: Global sti ness matrix entries where is the row multiplier.Formulation Finite Di erence Full-Integration (CM) Full-Integration (LM) Full-Integration (GM) Full-Integration (RC) Reduced-Integration (CM) Reduced-Integration (LM) Reduced-Integration (GM) Reduced-Integration (RC) Mm. Formulation Km. = vp=c. = vg =c. the non-dimensional phase speed. GM the generalized mass.n Km 1. LM the lumped mass.n 1 Table 1: Global Mass matrix entries where is the row multiplier. wave solution in equation (4) into equation (11). is 2 = 1 q f?g32 ? g4g 2 : x 2 (1 + ) ? f3 ? f4] 1 2) 2) (15) Remark Returning to the true group velocity. the propagation direction for the nite di erence semi-discretization is ( ) sin sin( k x)] : = arctan sin cos( k x)] (16) .

v. :::. Upon inspection of equation (14).For a unit aspect ratio. While equation (19) is identical to the non-dimensional phase speed presented in 3] for the full-integration bilinear element. . the projected group speed exactly represents the true group speed in the grid-aligned propagation directions. . Table 3 shows the functions required for use in equations (17) and (18) for the computation of the phase and group speed for all of the formulations considered in this investigation. 180. ( @v ) 1 v @u ? u @k r v : @k =2 x (18) v2 u Equations (17) and (18) are the general dispersion formulae for the non-dimensional phase and group speed for semi-discretizations of the second-order wave equation. M = Mc + (1 ? )Ml . f3. For non-unit aspect ratio. f3. k. Making use of the formulae in Table (3). 90. . Similar results may be shown for the nite element semi-discretizations. the group velocity aligns with the wave vector for = 0. k) and v = v(f1 . equation (20) yields phase speed curves identical to those in Figure 4 in 3] which con rms this form of the reduced-integration phase speed. Thus. By making use of equation (17). Nevertheless. x. the group speed may be written in terms of u and v. @u=@k. 135. and their derivatives with respect to the wave number. v u 1 u 6 (1 + 2 )(4 ? f1 ? f2 ) + 2( 2 ? 2)f3 + 2(?2 2 + 1)f4] (19) = k xt (1 ? )18 + 8 + f1 + f2 + 4(f3 + f4)] Similarly. the non-dimensional phase speed for the full-integration element may be written down in terms of the mass ratio. All that remains is to de ne the speci c u. (17) =k x v where u = u(f1. the non-dimensional phase speed for the reduced-integration element is v u 1 u 4 (1 + 2 )(2 ? f1 ? f2 ) ? 2(1 ? 2)f3 ? (2 2 ? 2)f4] : t (20) = k x (1 ? )8 + 2 + f1 + f2 + 2(f3 + f4)] Similar relationships for the non-dimensional phase speed are presented by Mullen and Belytschko 3]. f2. Equation (20) is considerably di erent from the phase-speed presented for the reduced-integration element. the non-dimensional phase speed may be written more generally and compactly as 1 ru. in the results that follow. = for grid-aligned wave propagation directions. f2. 45. . x. k). f4. and when j sin j = jcos j. . . and the phase speed formulae for the nite element formulations. f4. @v=@k relationships for each semidiscretization. Note that the functions presented in Table 3 have been developed for the generalized mass matrix. 360o.

The non-dimensional phase and group speed are presented as functions of the propagation angle. two grid aspect ratios are considered. 90o are included with each isometric plot to yield a more quantitative presentation. v. however. the non-dimensional wave number is interpreted as k x= = 2 x= where is the wavelength. 45. Thus. in order to demonstrate the interaction between propagation angle and varying grid resolution. Here. 67:5. 3 Results This section presents the results of the Fourier analysis in terms of the non-dimensional phase and group speed for the nite di erence and nite element semi-discretizations of the second-order wave equation. and non-dimensional wave number. 0 2 x= 1 represents the entire discrete wavelength spectrum with the Nyquist limit occurring at 2 x= = 1. @u=@k. In the ensuing discussion. All of the isometric plots are symmetric about = 90o.u = 2( 2 + 1) ? 2f3 ? 2 2f4 FDM v=1 @u=@k = ?2g3 ? 2 2g4 @v=@k = 0 u = 6 (1 + 2)(4 ? f1 ? f2 ) + 2( 2 ? 2)f3 + 2(1 ? 2 2 )f4] FEM v = f(1 ? )18 + 8 + f1 + f2 + 4(f3 + f4)]g Full-Integration @u=@k = 6 ?(1 + 2 )(g1 + g2) + 2( 2 ? 2)g3 + 2(?2 2 + 1)g4] @v=@k = g1 + g2 + 4(g3 + g4)] u = 3 (1 + 2)(4 ? f1 ? f2 ) + 2( 2 ? 2)f3 + 2(1 ? 2 2 )f4 FEM v = 5 + 2(f3 + f4) Full-Integration @u=@k = 6 ?(1 + 2 )(g1 + g2) + 2( 2 ? 2)g3 + 2(?2 2 + 1)g4] Reduced-Coupling @v=@k = 2(g3 + g4) u = 4 (1 + 2)(2 ? f1 ? f2 ) + 2( 2 ? 1)f3 + 2(1 ? 2 )f4] FEM v = (1 ? )8 + 2 + f1 + f2 + 2(f3 + f4 )] Reduced-Integration @u=@k = 4 ?(1 + 2 )(g1 + g2) + 2( 2 ? 1)g3 + 2(? 2 + 1)g4] @v=@k = g1 + g2 + 2(g3 + g4)] u = 2 (1 + 2)(2 ? f1 ? f2 ) + 2( 2 ? 1)f3 + 2(1 ? 2 )f4] FEM v = 2 + f 3 + f4 Reduced-Integration @u=@k = 2 ?(1 + 2 )(g1 + g2) + 2( 2 ? 1)g3 + 2(? 2 + 1)g4] Reduced-Coupling @v=@k = g3 + g4 Table 3: u. contour lines for = 0. In addition to the isometric plots. the results are plotted for 0 180o for the sake of clarity. a propagation angle of 0o corresponds . = 0:5 and = 1:0. @v=@k functions for computing the phase and group speed for the FDM and FEM formulations. 22:5. . For each semi-discrete formulation. The non-dimensional phase and group speed results are presented in the form of isometric plots for each of the semi-discrete formulations considered.

when the propagation direction is normal to the edges of a parallelogram that is constructed by connecting the outer nodes of the ve-point nite di erence stencil as shown in Figure 4. rapid changes in mesh resolution. As shown in Figure 3. when = 0o. 135o as shown in Figure 3(a-d). there are preferred propagation directions on the = 0:5 grid for which the phase and group errors are minimized for all wavelengths. the group speed for grid-aligned signals is zero indicating that spurious. Conversely. For = 0:5. the twodimensional solution consists of independent one-dimensional solutions for each y-grid line.e. while = 90o corresponds to a plane wave moving in the positive y-direction. 180o). only three of the ve nodes in the penta-diagonal operator actually contribute . propagation angles. For these propagation directions. non-dispersive physical wave propagation problem. it remains nite for all other propagation directions .to a plane wave moving down the positive x-axis. the phase and group errors are a minimum along = 45. Although the group speed goes to zero for 2 x wavelengths along = 0. i.. 180o. shown in Figure 3(e-h). and aspect ratios considered. = 1. or initial conditions with frequency content that exceeds the mesh Nyquist limit. in the = 90o direction. i. These results demonstrate that the nite di erence discretization introduces strictly lagging phase and group error for all wavelengths. In terms of the aspect ratio. the dispersive errors are larger for waves propagating along the coordinate directions. exhibit lagging errors that are not minimized in the direction of the higher resolution grid. the minimum phase and group errors occur. In this case. In fact. This type of short wavelength signal may be induced by boundary e ects. 90. K . i. in general. The non-dimensional phase and group speed for the nite di erence discretization are shown in Figure 3. the optimal propagation directions for 0 180o are given by = 90o 180 arctan : (21) opt 3. Thus. the phase and group speed depend not only on wavelength but also on the propagation direction with the angular variation increasing with 2 x= . short wavelength packets will be stationary on the grid. Here.e. the plane wave propagation problem is essentially one-dimensional. signals aligned with the grid ( = 0. For example.1 Finite Di erence Discretization The preferred propagation direction can also be interpreted in terms of how many o -diagonal nodes of the ve-point nite di erence sti ness operator. For the unit aspect ratio.e... the ideal phase and group speed should be independent of wavelength and propagation direction. are \actively coupled" in the solution for the propagating plane wave. the phase and group speed. Similar to the unit aspect ratio grid. the spatial discretization introduces anisotropy in addition to dispersion into an otherwise isotropic. For 2 x wavelengths. the phase and group speed remains nite for all wavelengths.undoubtedly a consequence of the higher resolution mesh in the y-direction. and the deviation of the non-dimensional phase and group error from unity is an indicator of the severity of the dispersive errors.

the numerical results show a 4 x wave superimposed over the left and right square waves with a 2 x wave centered at x = 1=2. It is still quite surprising that the best accuracy. In contrast to the results for the nite di erence discretization. Figure 5 shows several snapshots of a one-dimensional numerical experiment with a 20 x wide \top-hat" initial pro le. in the = 90o direction for = 0:5. For the unit aspect ratio grid. In other words. along the opt propagation direction. For nite . x for = 90o.2 FEM Discretization . For = 0:5.to the solution. the use of a consistent mass matrix with full integration yields strictly leading phase error and predominantly leading group error. the phase error for grid-aligned signals achieves a maximum at 2 x= 0:8. However. while the 4 x waves move at a nite speed albeit slower than the primary signal. To understand this. As predicted. c. This wave form was used to intentionally over-drive the grid resolution and introduce spurious. short wavelength oscillations. n) to node (m +1. Returning to the group error. the 2 x waves at x = 1=2 are stationary. the grid spacing from node (m. the ve-point nite di erence stencil appears to yield an e ective grid spacing of h = xj cos opt j in the opt propagation direction. In contrast. the group speed is zero for grid-aligned 2 x= wavelengths.. Waves propagating in the directions corresponding to opt in equation (21) experience the smallest leading phase and group errors up to 2 x= 0:8. beyond 2 x= = 0:8. rst consider that the e ective node spacing is p x for = 0o. Here. 3. Similarly. a higher resolution mesh. e. However. the minimum leading phase and group errors occur in the opt direction . in e ect. the group speed achieves a maximum at 2 x= 0:65. individual waves will appear to move faster in the opt direction than in any other direction on the grid. = 1:0. = 2 arctan as shown in Figure 4. This e ect is seen in Figure 3b and 3d where the phase and group speed for a 2 x wave along = 45o is identical to the phase and group speed along p = 0o at 2 x= = 1= 2. in terms of the phase and group speed. The exact solution for this problem consists of a left-moving and rightmoving square wave each with an amplitude. they serve as a representative example of the limiting case of grid-aligned wave propagation and the relationship between group speed and modulated 2 x waves. a 2 x= = 1 wave propagating in the opt direction sees.g. However. P = 1=2. and propagation speed. n +1) is greater than the grid-aligned node spacing. Similar to the nite di erence results. Thus. the phase and group speed for the unit aspect ratio grid achieve a maximum for 2 x waves in the opt directions. are shown in Figure 6. the plane wave solution relies upon information from all of the nodes in the ve-point nite di erence operator yielding the reduced phase and group errors. Although the results in Figure 5 are from a one-dimensional computation. 135o. does not occur in the direction of highest mesh resolution.Full Integration Element The non-dimensional phase and group speed results for the full-integration nite element formulation with a consistent mass matrix. and x 1 + 2 sin for = opt . and d=2 in Figure 4. while these wave packets propagate with a group speed greater than the sound speed for = 45.

is shorterp the distance between nodes (m + 1. viz. the optimal propagation direction does not correspond to Equation (21). the optimal propagation directions coincide with the grid directions for the unit aspect ratio. i. nine-point coupling of the time derivative terms in equation (12) and leads to strictly lagging phase and group errors for the fully integrated element as shown in Figure 7. This is shown more clearly in Figure 8a and b where the e ect of the mass lumping on the group error is a small. for the unit aspect ratio. This can be seen by considering the ratio of a 2 x wavelength.. . n). However. 135o in Figure 4a and c. n) as shown in Figure 4. 2 x=d = 1:12 for = 0:5. In comparison to the nite di erence results. 180o as shown in Figure 7(a-d). In the preferred propagation directions for the unit aspect ratio grid. negative group speed for = 45. For = 1:0. the propagation angles that yield minimum errors in phase and group speed for this element correspond approximately to a wave propagation direction normal to the edges of the bounding parallelogram drawn for a ve-point stencil centered at node (m. In general. n + 1) than and (m ? 1.e. This in uence is re ected by the increased phase and group error for = 45. Lumped Mass The process of row-sum lumping the nite element mass matrix yields a diagonal mass matrix. n). mass lumping yields slightly larger lagging phase and group errors except in the grid-aligned propagation directions. This removes the broad. However. for = 0:5. The nite group speed associated with = 90o is simply a consequence of the higher resolution mesh in this direction. a 2 x wave appears to have a 4 x wavelength for = 90o. and the characteristic dimension of the 4-patch of elements surrounding node (m.e. 2 x=d = 1= 2. The consequence being that the consistent mass and sti ness operators can couple information that is outside the wavelength of the propagating signal. the phase and group error is always lagging. i. centered at node (m. a 2 x signal centered about node (m. n). the 2 x wave encompasses all nine nodes of the nite element sti ness operator. and the higher resolution mesh bias for non-unit aspect ratio. the e ect of the lumped mass is to make the nite element method behave more like the nite di erence method. 135o. K . i. 90.e. the group speed is slightly negative at the Nyquist grid limit indicating that 2 x wave packets will propagate backwards relative to the longer wavelength components of a signal. 90o. n ? 1). In this case. unlike the nite di erence semi-discretization.. the optimal propagation angle is 90o which corresponds to the high resolution bias for the mesh.according to equation (21). In Figure 7d. For = 0:5. Thus. the group speed goes to zero in the limit of 2 x wavelengths for = 0. Instead.. the optimal propagation directions occur for = 0. (22) d 1 + 2 sin where = 2 arctan as shown in Figure 4. This ratio is 2 x=p 1 .

Similar results may also be shown for = 90o. In addition. v u 1 u 2 2 ? cos(k x)] : = k x t 1 + cos(k x) (23) Thus. The singular behavior of the phase speed may be understood by considering the limiting case of = 0o.Higher-Order Mass The concept of a higher-order mass matrix has appeared in the work of Goudreau 24]. the contribution of the lumped mass produces a signi cant e ect in terms of the lagging phase and group error. Note that this element is usually implemented with a lumped mass matrix because of the poor dispersion characteristics associated with its consistent mass matrix. the results in Figure 10(a-d) show that the phase and group speed become unbounded for 2 x waves at = 0. = 1. MacNeal 25] and Belytschko and Mullen 5]where improved accuracy in terms of phase speed has been obtained by using a linear combination of the lumped and consistent mass matrices.e. for the group speed.3 FEM Discretization . = 1. For = 0:5. in the limit as k x ! . The dispersion curves for the fully integrated element with higher-order mass are presented in Figure 9.e. 180o.Reduced Integration Element . the higher-order mass matrix reduces the mesh requirements relative to the lumped mass matrix by a factor of two in a one-dimensional sense. 3. as 2 x= ! 1. the phase and group error become unbounded for 2 x waves propagating along = 0. Figure 10 shows the non-dimensional phase and group speed results for this element. Because cosine is an even function. For the unit aspect ratio grid. 180o. i. the phase and group error plots are relatively at for the shorter wavelengths and intermediate angles.. the contribution of the lumped mass leads to lagging phase and group error for all aspect ratios and propagation angles.. The reduced integration element is of interest because it can be implemented in a manner that requires minimal memory and computational cost. For a xed 2:5% phase error. Here. the leading group error for this element makes its use feasible only for 2 x= 0:2. or alternatively. The nite phase and group speed in the = 90o direction is clearly a consequence of the higher resolution grid. In this element. Selecting the appropriate u and v functions for this element from Table 3. this is also true for = 180o. the higher-order mass matrix is taken to mean that = 0:5 in Equation (3). For this element. the phase speed becomes unbounded. the higher-order mass matrix yields a marked reduction in both the phase and group error. However. The optimal propagation directions for the higher-order mass correspond to those for the lumped mass discussed above. as well as. 90. and making use of Equation (17). nearly a 15% improvement relative to the lumped mass for 2 x waves with = 1:0. a factor of four reduction in two-dimensions. While the phase speed for this element is relatively well behaved for 2 x= 0:5. as well as the angular sensitivity. ! 1. i. highly re ned meshes.

Row-sum lumping the corner nodes of the 3. The primary e ect of the row-sum lumped mass contribution to the higher-order mass matrix is to remove the singular behavior of the phase and group speed in the limit of 2 x wavelengths. while. n 1). wave packets p with 2 x= 1= 2 along = 45. Similar to the fully integrated element. 135o will move backwards on the grid. particularly for the reduced-integration element. Again. the phase and group error is reduced relative to both the consistent mass and lumped mass versions of this element. 2 x= wavelength packets will be stationary on the grid. there is a small leading phase error that may be attributed to the singular behavior of the consistent mass matrix. the group speed is zero at = 0. 90. although in general this element has slightly larger phase and group error than the corresponding nite di erence discretization. the largest dispersive errors occur along = 45. Mass lumping also removes the singular behavior for the 2 x waves observed with the consistent mass matrix. this element produces a negative group speed as indicated in Figure 11c and shown in Figure 12. In contrast.g. node (m 1. Similar results are shown in Figure 11(e-h) for = 0:5. the higher-order mass matrix for the reduced-integration element exhibits both leading and lagging phase and group error as shown in Figure 14.4 Reduced-Coupling Mass . For the unit aspect ratio grid. 180o for the unit aspect ratio grid. The optimal propagation angles for this element are identical to the full-integration element with lumped mass. Similar to the nite di erence results. the group speed is negative and nearly 60% of the true sound speed along = 45. to the broad spatial coupling associated with the consistent mass matrix.Lumped Mass The row-sum lumped mass matrix meliorates many of the pathologies associated with the reduced-integration. 135o.. e. As shown in Figure 14c and d. the group speed remains positive and bounded. The row-sum lumping procedure introduces strictly lagging phase and group error as illustrated in Figure 11. Again. it was observed that the corner nodes. 135o. in part. this element formulation yields minimum phase and group errors in the direction of the highest resolution mesh. 135o with negative group speed for 2 x= wavelength packets. from the nite element consistent mass matrix have a signi cant e ect upon the phase and group error. this element yields the smallest dispersive errors for propagation directions associated with the high-resolution bias in the mesh. In fact. Figure 15 shows expanded plots of the non-dimensional group speed and reveals that the group speed is negative and approximately 75% of the true sound speed for = 45. Higher-Order Mass Unlike the fully integrated version of this element. for = 0:5. An expanded view of the nondimensional group speed is shown in Figure 13. For all of the aspect ratios considered. 0 1. In the discussion above. Like its full-integration counterpart. consistent-mass quadrilateral element. This observation led to the hypothesis that the leading phase error may be due.

the so-called higher-order mass matrix. While the e ect of the reduced-coupling mass matrix is similar to that observed for the fully integrated element. in general the dispersion errors are improved relative to the higher-order mass matrix. 135o and = 1:0 as is the case for the higher-order form of the the reduced integration element. However. All of the discretization methods considered herein exhibit optimal propagation directions. In contrast.consistent mass results in a mass matrix that spatially couples the P terms of Equation (2) in the form of a ve-point nite di erence stencil. As shown in Figure 16. the behavior of the phase speed for 2 x waves and 1:0 is improved markedly. consider that a 2:5% phase error is admissible for . 90. the lumped mass matrix introduces strictly lagging phase and group error. the sensitivity of the dispersion error to propagation angle is reduced for 2 x= 0:8 making the identi cation of an optimal propagation direction impossible. This form of the mass matrix is termed the \reduced-coupling" mass matrix. 4 Summary and Conclusions The results of this analysis have demonstrated that the consistent mass matrix introduces strictly leading phase and predominantly leading group error. However. That is. as well as directional sensitivity that is due to the broad. While this element exhibits both lagging and leading phase and group error. This is evident in the leading phase and group error for = 0.d). The lagging phase errors that are introduced are not signi cant until 2 x= 0:8. the reduced-coupling mass matrix removes most of the leading phase and group error associated with the consistent mass matrix for the full-integration element. nine-point coupling of the time derivatives. i. Further. However. A linear combination of the lumped and consistent mass matrix. results in enhanced phase and group speed. in comparison to the behavior of the consistent mass matrix with reduced integration. Figure 17 illustrates the phase and group error for the reduced-integration element with the reduced-coupling mass matrix. beyond this limit. the positive in uence of the highresolution mesh on both the phase and group speed seems to be broadened with respect to propagation angle as shown in Figure 16(e-h). 180o as shown in Figures 17(a . and nearly non-existent for the full-integration element.e. the phase error is no longer zero for this wavelength for = 45. propagation directions that yield minimal phase and group error. for = 1. this e ect is minimized by the reduced-coupling mass for the reduced-integration element.. the reduced-integration element shows more sensitivity to propagation angle. the variation in the group speed with respect to propagation seems to be only slightly improved with respect to the full-integration element with a consistent mass matrix. this element is quite well behaved for virtually all wavelengths and propagation directions. In order to estimate the computational cost associated with the higher-order mass or the reduced-coupling mass matrix. In speci c. Conversely. but still su ers from the angular sensitivity of the broad spatial coupling attributed to the consistent mass. The reduced-coupling mass matrix enhances both the phase and group speed while removing the angular sensitivity of the consistent mass matrix.

a factor of nearly 2 reduction in grid requirements results with the reduced-integration element when either the higher-order or reduced-coupling mass matrix is used. Thus. This translates into a factor of four in two-dimensions { a non-trivial savings in grid requirements. the phase and group error for nite element discretizations are.. a factor of nearly two savings in grid resolution may be obtained using the higher-order mass matrix instead of the lumped mass with the full-integration element. the results presented here will provide guidance in terms of code design.engineering applications. and higher-order mass matrices.e. mesh aspect ratio. the reduced-coupling mass matrix yields dispersion results that are somewhat less sensitive to propagation direction than the higher-order mass matrix. the reduced-coupling mass matrix o ers both enhanced dispersion characteristics and reduced computational requirements for a speci ed level of accuracy. and propagation angle on the phase and group errors for the second-order wave equation. The results of this analysis demonstrate that in terms of angular dependence. in general. Table 4 presents the mesh requirements and sense of phase error. i. The use of a higher-order mass matrix can signi cantly reduce the phase and group error for both the full-integration and reducedintegration bilinear element. A factor of 16 reduction in the grid requirements clearly translates into very real savings in terms of storage and computational cost. Alternatively. In an average sense. associated with each of the discretization methods for a xed 2:5% phase error. a factor of nearly four savings in grid resolution may be realized with the reduced-coupling mass matrix for the full-integration element (a factor of 16 in two-dimensions). The reduced-coupling mass delivers superior phase and group speed with respect to the consistent. superior to a secondorder accurate nite di erence discretization. Similar to the full-integration element. This analysis has considered the in uence of discretization. The range in mesh resolution is due to the variation in the phase error with respect to propagation angle. Formulation Integration Mass Matrix Elements per Wavelength Finite Di erence N/A Diagonal 6 ? 8 (lagging) Finite Element Full Consistent ( = 1) 6 ? 8 (leading) Finite Element Full Lumped ( = 0) 8 ? 10 (lagging) Finite Element Full Higher-Order ( = 1=2) 4 ? 6 (lagging) Finite Element Full Reduced-Coupling 2 ? 3 (lagging) Finite Element Reduced Consistent ( = 1) 10 ? 14 (leading) Finite Element Reduced Lumped ( = 0) 8 ? 12 (lagging) Finite Element Reduced Higher-Order ( = 1=2) 6 (leading) Finite Element Reduced Reduced-Coupling 6 (leading) Table 4: Grid requirements to for a xed 2:5% phase error for Finite Di erence and Finite Element Formulations. imple- . lumped. While. but with a real cost in terms of storage. However. leading or lagging.

and mesh design for analysis. it remains to demonstrate the performance of the higher-order and reduced-coupling mass matrices in practical computations. Acknowledgments I wish to thank Professor Ted Belytschko at Northwestern University for encouraging me to document this work.mentation. Appendix A Fully-Integrated Quadrilateral Element Consistent Mass Matrix: 2 4 2 1 2 26 2 1 6 me = 36x 6 sym: 4 4 2 c 4 4 21 26 6 me = 4x 6 1 1 l 4 3 7 7 7 5 3 7 7 7 5 3 7 7 7 5 3 7 7 7 5 (24) Lumped Mass Matrix: (25) Reduced-Coupling Mass Matrix: 1 Sti ness Matrix: 2 5 2 0 2 26 2 0 me = 36x 6 sym: 5 5 2 6 rc 4 5 (26) 2 (2 2 + 2) (?2 2 + 1) (? 2 ? 1) ( 2 ? 2) 2 c2 6 ( 2 ?2 ke = 6 6 sym: (2 + 2) (2 2? 2) ((?2 2? 1) 6 4 + 2) + 1) 2 + 2) (2 (27) . and Randy Weatherby for their helpful suggestions during the preparation of this manuscript. Allen Robinson. I would also like to acknowledge Dave Gartling. John Shadid.

Fourier Analysis of Numerical Approximations of Hyperbolic Equations. Elsevier Science Publishers. Schreyer. Philadelphia. 18:1{29. Dispersion analysis of nite element semidiscretizations of the two-dimensional wave equation. PA. November 1994. SIAM. Wineman. In High Performance Computing in Computational Dynamics. International Journal for Numerical Methods in Engineering. Gerald L. pages 267{298. ASME. 2] Robert Vichnevetsky and John B. 3] Robert Mullen and Ted Belytschko. L. Sarah J.Reduced-Integration Quadrilateral Element Consistent Mass Matrix: me = c Lumped Mass Matrix: 2 x2 6 6 4 16 6 1 1 13 1 1 17 7 5 sym: 1 17 1 1 1 1 3 7 7 7 5 (28) me = l Reduced-Coupling Mass Matrix: 2 x2 6 6 4 4 6 2 x2 6 6 6 16 4 1 (29) 1 me = rc Sti ness Matrix: 2 6 6 6 4 1 0 13 2 1 07 7 5 sym: 2 17 2 2 3 7 7 7 5 (30) c2 ke = 6 ( 2 + 1) (? 2 + 1) (? 2 ? 1) ( 2 ? 1) ( 2 + 1) ( 2 ? 1) (? 2 ? 1) sym: ( 2 + 1) (? 2 + 1) ( 2 + 1) (31) References 1] Mark A. pages 25{38. 4] H. (LLNL UCRL-JC-117820). 1982. chapter Dispersion of semi discretized and fully discretized systems. Bowles. and James D. 1983. Computational Methods for Transient Analysis. 1982. Foch. . A mixed time integration method for large scale acoustic uid-structure interaction. Christon. volume 1. Goudreau.

15] K. A fourier analysis of spurios mechanisms and locking in the nite element method. 7:409{ 452. June 1975. 128. Bethlehem. 14] K. R. Park and D. A new nite element formulation for computational uid dynamics: Ix. Flaggs. 11(1-3):67{78. Computers and Mathematics with Applications. 1978. Applied Numerical Mathematics. Achenbach. 13] K. 12] Smadar Karni. 6] Rober Vichnevetsky and A. 24(2):113{136. 9] Robert Vichnevetsky and E. C. Princeton University. SIAM Review. 1985. Flaggs. Pennsylvania.5] Ted Belytschko and Robert Mullen. C. Spurious wave phenomena in numerical approximations of hyperbolic equaitons. On the group velocity of symmetric and upwind numerical schemes. Pariser. 1987. Tomalesky. . Computer Methods in Applied Mechanics and Engineering. Park. International Union of Theoretical and Applied Mechanics. W. 7] R. 10:195{229. Wave propagation analysis of di erence schemes for hyperbolic equations: a review. F. In Proccedings of the Fifth Annual Princeton Conference on Information Science and Systems. Hughes. Vichnevetsky and F. In Julius Milovitz and Jan D. pages 67{82. Lehigh University. International Journal for Numerical Methods in Fluids. On dispersive properties of nite element solutions. pages 117{128. 11] Lloyd N. 1985. 1994. On wave propagation in almost periodic structures. 18:1073{1081. AMD-Vol. In Structural Acoustics. Trefethen. Computer Methods in Applied Mechanics and Engineering. John Wiley and Sons. pages 179{185. Park and D. High order numerical sommerfeld boundary conditions: theory and experiments. Frequency-window tailoring of nite element models for vibration and acoustics analysis. 1992. editors. 1991. Alvin and K. A symbolic forier synthesis of a one-point integrated quadrilateral plate element. Group velocity in nite di erence schemes. Computer Methods in Applied Mechanics and Engineering. 16] Farzin Shakib and Thomas J. 1984. International Journal for Numerical Methods in Fluids. 46:65{81. L. L. March 1971. 8] Robert Vichnevetsky. Modern Problems in Elastic Wave Propagation. pages 357{363. 87:35{58. 48:203{236. A frequency analysis of nite di erence and nite element methods for initial value problems. fourier analysis of space-time galerkin/least-squares algorithms. ASME. C. C. 1991. April 1982. In AICA International Symposium on Computer Methods for Partial Di erential Equations. 10] Robert Vichnevetsky. De Schutter.

Hughes. R. Technical Report NASA SP-222(01). Computer Methods in Applied Mechanics and Engineering. Computer Methods in Applied Mechanics and Engineering. 1972. Hot Rinehart and Winston. Inc. 1994. Civil Engineering Department. The nastran theoretical manual. Mathematical Analysis of Physical Problems.. Complex wave-number dispersion analysis of galerkin and galerkin least-squares methods for uid-loaded plates. Technical Report UCRL-MA-114536. 13:255{275. R. 18] Michel O. Christon. Complex wavenumber fourier analysis of the p-version nite element method. Berkeley. R. March 1992. 5. 24] Gerald L. May 1993. 5. Evaluation of discrete methods for linear dynamic response of elastic and viscoelastic solids. 1994. H. Englewood Cli s. Prentice-Hall. 21] P.5-3.17] Isaac Harari and Thomas J. Pinsky. 113:67{98. MacNeal. Hughes. Ping: An explicit nite element code for linear structural acoustics . New York. level 15. 98:411{454. SIAM Journal for Scienti c and Statistical Computing. 19] Lonny L. Wallace. The Finite Element Method. .5-4. 1992. 1987. New York. Computational Mechanics. New Jersey.5. NASA. University of California. Pinsky. Thompson and Peter M. 20] Karl Grosh and Peter M. 13(2):596{610. Lawrence Livermore National Laboratory. Deville and Ernest H. Mund. 25] R. 1972. 22] Mark A. PhD thesis.user manual. Fourier analysis of nite element preconditioned collocation schemes. 23] Thomas J. 1970. Galerkin/least-squares nite element methods for the reduced wave equation with non-re ecting boundary conditions in unbounded domains. Goudreau.

n n1 n4 2 n1 m-1.n+1 n3 3 n2 m+1.n+1 ∆y m-1.n-1 m+1.n ∆x m. (b) on a nite di erence grid.∆x m-1.n n3 m+1.n m.n-1 n2 n1 m.n+1 m+1.n-1 (a) (b) Figure 1: Propagation direction: (a) on a 2x2 patch of a nite element mesh.n). m-1. .n-1 m.n+1 n3 n4 4 m.n-1 m-1.n+1 θ m.n m-1.n-1 Figure 2: Four element patch with local node numbers for the nite element assembly of a global equation at node (m.n-1 n2 n1 n3 n4 m.n θ m+1.n x m.n n2 m+1.n+1 n4 1 m-1.n m.n+1 ∆y x m+1.

and = 0:5 in (e)-(h). = 1:0 in .(a) (b) (c) (d) (e) (f) (g) (h) Figure 3: Non-dimensional nite di erence phase ( ) and group speed ( ): (a)-(d).

n-1 m+1. .n m+1.n+1 Five-point Stencil Bounding Parallelogram θ m-1.n m-1.n-1 h Figure 4: Grid section showing optimal propagation direction relative to the bounding parallelogram for a ve-point nite di erence stencil.n-1 m.n+1 β m+1.n m.d m-1.n+1 ∆y ∆x m.

1:2 1:0 0:8 0:6 P 0:4 0:2 0:0 ?0:2 0:0 a) 1:2 1:0 0:8 0:6 P 0:4 0:2 0:0 ?0:2 0:0 1:2 1:0 0:8 0:6 P 0:4 0:2 0:0 ?0:2 0:0 9 Initial Pro le 0:1 0:2 0:3 0:4 0:5 x 0:6 0:7 0:8 0:9 1:0 ) 4 x wavelength 2 x wavelength N b) 0:1 0:2 0:3 0:4 0:5 x 0:6 0:7 0:8 0:9 1:0 2 x wavelength N 0:1 0:2 0:3 0:4 0:5 x 0:6 0:7 0:8 0:9 1:0 c) Figure 5: Snapshots of wave propagation along x-axis: (a) initial 20 x pro le and solution at t = 15 x=c. . (b) solution at t = 25 x=c. and (c) solution at t = 35 x=c.

and = 0:5 in (e) . .(h).(a) (b) (c) (d) (e) (f) (g) (h) Figure 6: Non-dimensional nite element phase ( ) and group speed ( ) for the fullintegration element with consistent mass matrix ( = 1:0): = 1:0 in (a) .(d).

(d).(a) (b) (c) (d) (e) (f) (g) (h) Figure 7: Non-dimensional nite element phase ( ) and group speed ( ) for the fullintegration element with lumped mass matrix ( = 0:0): = 1:0 in (a) .(h). . and = 0:5 in (e) .

(a) (b) Figure 8: Non-dimensional nite element group speed ( ) for the full-integration element with lumped mass matrix ( = 0:0) and unit aspect ratio ( = 1:0). .

(d). .(a) (b) (c) (d) (e) (f) (g) (h) Figure 9: Non-dimensional nite element phase ( ) and group speed ( ) for the fullintegration element with higher-order mass matrix ( = 0:5): = 1:0 in (a) .(h). and = 0:5 in (e) .

and = 0:5 in (e) .(d).(h). .(a) (b) (c) (d) (e) (f) (g) (h) Figure 10: Non-dimensional nite element phase ( ) and group speed ( ) for the reducedintegration element with consistent mass matrix ( = 1:0): = 1:0 in (a) .

(d).(h).(a) (b) (c) (d) (e) (f) (g) (h) Figure 11: Non-dimensional nite element phase ( ) and group speed ( ) for the reducedintegration element with lumped mass matrix ( = 0:0): = 1:0 in (a) . . and = 0:5 in (e) .

.(a) (b) Figure 12: Non-dimensional nite element group speed ( ) for the reduced-integration element with lumped mass matrix ( = 0:0) and unit aspect ratio ( = 1:0).

(a) (b) Figure 13: Non-dimensional nite element group speed ( ) for the reduced-integration element with lumped mass matrix ( = 0:0) and = 0:5. .

(d).(a) (b) (c) (d) (e) (f) (g) (h) Figure 14: Non-dimensional nite element phase ( ) and group speed ( ) for the reducedintegration element with higher-order mass matrix ( = 0:5): = 1:0 in (a) . .(h). and = 0:5 in (e) .

.(a) (b) Figure 15: Non-dimensional nite element group speed ( ) for the reduced-integration element with higher-order mass matrix ( = 0:5) and unit aspect ratio ( = 1:0).

. and = 0:5 in (e) .(d).(h).(a) (b) (c) (d) (e) (f) (g) (h) Figure 16: Non-dimensional nite element phase ( ) and group speed ( ) for the fullintegration element with the reduced-coupling mass matrix: = 1:0 in (a) .

(d). and = 0:5 in (e) .(a) (b) (c) (d) (e) (f) (g) (h) Figure 17: Non-dimensional nite element phase ( ) and group speed ( ) for the reducedintegration element with the reduced-coupling mass matrix: = 1:0 in (a) . .(h).

- Terminal 3 Map2_2015_R3_tcm13-4102
- Static Head Calculation for B-31.docx
- Export
- Vibration Related Failures of Small Bore-Attachments (1)
- Synthron_Final_Report1.pdf
- Talk
- Sphere MAWP
- LWN Tables
- 510_BOK_May-2015-to-Jan-2016_Final_120414.pdf
- 503690-BonusChapter02.pdf
- a-18 top head
- Pressure Vessel MI Field Template
- Graduation Ticket Vipin
- CF Gate Valve GT80_data_sheet
- test1
- 05 Minor Moc Sdr Form
- Redlined Dwg (2)
- PVElite Installation Method.txt
- e Battery Tanks
- 15CSTL-NA
- 30S316-H2S
- 60CSTL-H2S
- 30S304-NS
- Roll Out and Maintenance Integration of SIS Proof Test and Inspection_Rev 0
- 514978-Updated Iah Terminal Map Draft 7-17-13

Sign up to vote on this title

UsefulNot usefulNFEA

NFEA

- Transfer matrix
- sasw1.pdf
- PH 333 Lecture 92014-2
- The Path Difference
- waves excercise
- aa
- Ganan-V1,3 13
- 6734_01_que_20090121
- D23_assessmentreport_waveguideFEM
- Microwave Experiment
- PHYS632_C15_35_Interf
- Waves and Light Unit Exam Practice Questions.doc
- CE Phy Mock Paper A2 Q& A
- tb_chapter13.pdf
- Science of Waves
- Wave Interactions
- Vector 1
- Huygens Snell Fermat Seismic Reflection
- m205
- Principles of Quantam Mechanics
- NS101_Lab3_Spring2012
- p1 Rev Resources
- 4 Introduction to Waves
- Phys2CL_Exp4_2010
- Wave-particle Duality on Macro Scale
- Ne3AChap
- 8_oscillations_waves_2.pdf
- Taller Mas Fisica 3
- (E-book CFD) - Implementation of Finite Element-Based
- Tutorial5
- NFEA