论文李新亮的共享数据程序、文档及课件论文hezw-2014_W_第1页
论文李新亮的共享数据程序、文档及课件论文hezw-2014_W_第2页
论文李新亮的共享数据程序、文档及课件论文hezw-2014_W_第3页
论文李新亮的共享数据程序、文档及课件论文hezw-2014_W_第4页
论文李新亮的共享数据程序、文档及课件论文hezw-2014_W_第5页
已阅读5页,还剩6页未读 继续免费阅读

下载本文档

版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领

文档简介

1、SCIENCE CHINAPhysics, Mechanics & Astronomy Article April 2014 Vol.57 No.4: 753763doi: 10.1007/s11433-013-5234-yNonlinear spectral-like schemes for hybrid schemesHE ZhiWei1, LI XinLiang2* & LIANG Xian21 China North Vehicle Research Institute, Beijing 100072, China;2 State Key Laboratory of High Temp

2、erature Gas Dynamics, Institute of Mechanics, Chinese Academy of Sciences, Beijing 100190, ChinaReceived September 26, 2012; accepted March 26, 2013; published online February 19, 2014In spectral-like resolution-WENO hybrid schemes, if the switch function takes more grid points as discontinuity poin

3、ts, the WENO scheme is often turned on, and the numerical solutions may be too dissipative. Conversely, if the switch function takes less grid points as discontinuity points, the hybrid schemes usually are found to produce oscillatory solutions or just to be unstable. Even if the switch function tak

4、es less grid points as discontinuity points, the final hybrid scheme is inclined to be more stable, provided the spectral-like resolution scheme in the hybrid scheme has moderate shock-capturing capability. Fol- lowing this idea, we propose nonlinear spectral-like schemes named weighted group veloci

5、ty control (WGVC) schemes. These schemes show not only high-resolution for short waves but also moderate shock capturing capability. Then a new class of hy- brid schemes is designed in which the WGVC scheme is used in smooth regions and the WENO scheme is used to capture discontinuities. These hybri

6、d schemes show good resolution for small-scales structures and fine shock-capturing capabilities while the switch function takes less grid points as discontinuity points. The seven-order WGVC-WENO scheme has also been applied successfully to the direct numerical simulation of oblique shock wave-turb

7、ulent boundary layer interaction.group velocity control, hybrid, WENO, shockwave-turbulent boundary layer interaction, direct numerical simulation PACS number(s): 47.11.Bc, 47.40.Nm, 47.27.ekCitation:He Z W, Li X L, Liang X. Nonlinear spectral-like schemes for hybrid schemes. Sci China-Phys Mech Ast

8、ron, 2014, 57: 753-763, doi: 10.1007/s11433-013-5234-yShock wave-turbulent boundary layer interaction (SWTBLI) problems, characterized by turbulences and shockwaves, are archetypical examples of multi-scales problems with dis- continuities. Because of the importance and widely applica- tions of thes

9、e problems 1, direct numerical simulations (DNS) of SWTBLI problems have been required continual- ly.The multi-scales phenomena in such problems requires the numerical method to be high order and high resolution. To be reliable, a DNS of such flows must resolve these var- ious scales, particularly s

10、maller ones with accuracy in both amplitude and phase 2. Concerning high-order and high- resolution schemes, there have been upwind compact schemes 3,4, dispersion-relation preserving schemes 5*Corresponding author (email: )and wavenumber-extended finite difference schemes 6. These sc

11、hemes are all designed for high resolution of short waves with respect to the computational grid. Another high- resolution schemes are explicit/compact central schemes 7 which have no dissipation resulted in providing spurious solutions and leading to inevitable stability problems 8,9.However, these

12、 methods above are limited to compute flows without discontinuities. To capture shocks, treating these methods nonlinearly is required 4. For flows with discontinuities, WENO 10 schemes which have been demonstrated very promising shock-capturing capabilities and high order accuracy are widely used.

13、However, numer- ical tests also indicate that classical WENO schemes are usually not optimal for computing turbulent flows or aero- acoustic fields because they are too diffusive for short waves 11. Science China Press and Springer-Verlag Berlin Heidelberg 2014763He

14、 Z W, et al.Sci China-Phys Mech AstronApril (2014) Vol. 57 No. 4The observations presented above suggest that, a natural choice is to combine the ENO/WENO scheme with another scheme with spectral-like resolution to form a so-called spectral-like resolution-WENO hybrid scheme 2. Adams and Shariff 12

15、have developed hybrid schemes that rely on the coupling of a nonconservative compact upwind scheme with a shock-capturing ENO scheme that is turned on around discontinuities. Pirozzoli 13, following ref. 12, derived a hybrid compact-WENO scheme in which a con- servative compact scheme was coupled wi

16、th a WENO scheme to make the overall scheme conservative. Ren et al.11 improved the hybrid compact-WENO scheme 11 by removing the abrupt switch between the compact and the shock-capturing schemes through the use of a weighted av- erage of the two schemes. There are still other spectral-like resoluti

17、on-WENO hybrid schemes 2,14.However, there are still problems with switch functions in these hybrid schemes. Take the switch function devel- oped by Ren et al. 11 as an example. It was found that this switch function may mistake some points in smooth waves as discontinuity points, particularly in hi

18、gh-wavenumberto capture discontinuities. The final hybrid schemes show good resolution for small-scales structures and fine shock- capturing capabilities. And the seven-order WGVC-WENO scheme has also been applied successfully to the DNS of oblique SWTBLI.1 Group velocity of linear finite difference

19、 schemes and WENO schemesTrefethen 19 surveyed and illustrated the relevance of group velocity to the behavior of finite difference models of time-partial differential equations. For more details on the WENO scheme, refer elsewhere 10. Here we briefly list these common sense knowledge for completene

20、ss.1.1 Group velocity of linear finite difference schemesConsider the one-dimensional linear advection equation on an infinite domain with a sinusoidal initial condition with wavenumber k ,waves 15. Thus a problem arises. That is, if the switchfunction takes more grid points as discontinuity points,

21、 theu + c u= 0, - x 0.Theoretically, applying separation of variables, it is clear that the exact solution to eq. (1) isu(x,t) = u0eik (x-ct).(2)In practice, eq. (1) is discretized in the spatial domain on a uniform grid with nodes given by xj=jDx, j = 0, N ,where Dx is the cell size. To facilitate

22、the discussion, let operator L denote the approximation of the spatial differen-17 theory. This concept is proposed and used to explain the phenomenon of oscillations near the discontinuities and capture discontinuities 16,17. The main ideas are that atial, L(u j ) ujDxx x= x. Here we consider an ex

23、plicit, lineardiscontinuity can be treated as a linear combination of long and short waves. Oscillation productions in numerical solu- tion around discontinuities are due to non-uniform groupvelocity of wavelets. Thus to control group velocities of wavelets could be used to reduce oscillation produc

24、tionsfinite difference operator, that is,srL(u j ) = alu j +l ,l =-sl(3)around discontinuities. GVC schemes does be found nu- merically to have a ability of reducing many oscillations near discontinuities while there still a bit left 17,18.In this paper, firstly, by following the GVC theory, wewhere

25、 sl and sr are arbitrary positive integers and al, l = -sl ,L, sr are constant coefficients. Then the semi- discretized form of eq. (1) by the method of lines yields a system of ordinary differential equationspropose a weighted methodology of constructing GVC sch-du jL(u j )emes. A new class of sche

26、mes named weighted group ve- + c = 0, u(0) = u eik xj , j = 0,., N,(4)locity control (WGVC) schemes is deduced. Then we pay our main attention to show WGVC schemes have high-dtDx0where u =u(x , t). Then the exact solution to theorder and high-resolution property, not to demonstrate theirjjmoderate s

27、hock-capturing capabilities. Finally, a new class of hybrid schemes is designed in which the WGVC schemesemi-discrete equation isu(x ,t) = u ei(k x j -Kect),(5)is used in smooth regions while the WENO scheme is usedj0where Ke (k) = 1iDxsr al el =-sli(lDx). For convenience we in-anda = Ck,(9)ktroduce

28、 the reduced wavenumberx = kDx 7 and the re-(bk + e ) pduced modified wavenumber X(x ) = Ke (k) Dxwhich is aw =ak.(10)kk =0complex number Re(X)+ i Im (X). Then we can rewritekr -1 aeq. (5) as:j0u(x ,t) = ueIm(X)ct /Dxeik ( x j -Re(X)ct /x ) .(6)The small positive number e prevents division by zero,

29、and p is used to increase or decrease WENO adaptationIm(X) contains information on the spectral dissipationsensitivity. Details can be found elsewhere 10,20. property of the scheme. Re(X) contains information onthe spectral dispersion property of the scheme. Details can be found in refs. 2,4,19.In r

30、ef. 16, group velocity of a this linear difference2 WGVC2.1 Two base schemes: mixed and slow operatorxscheme is definedd (Re(X) )dfor constant c. And a newWe need a fast/mixed scheme and a slow scheme as twoclassification for the linear difference operator L is intro-dbase schemes which are used in

31、the two sides of a disconti-nuity respectively.duced, that is, the operator L: withdx (Re(X) 1,In this paper, we use the following 2(r-1) order operator with a free parameter0 x pis a slow operator, denoted by Ls; with d (Re(X) 1 , 0 x p is a fast operator; denoted by dxL( f j ;s ) = s f j-r +r -1l

32、=-(r -1)al (s ) f j+l(11)L ; with$p x 0 , 0 x x, d (Re(X) ) 1to obtain the mixed or slow operator. And its numerical fluxandf0x d0dxapproaches hj+1/2=h(xj+1/2) isr -1x0 x p,ed by Lm.(Re(X) ) 1dis a mixed operator, denot-f j +1/2 (s )=l =-(r -1)bl (s ) f j +l ,(12)According to the classification, a f

33、ast operator tends towhere the h(x) is implicitly defined as in ref. 21,propagate a wave in a faster speed than the correct speed in the semi-discrete evolution. A slow operator tends to prop- agate a wave in a slower speed than the correct speed in thef (x) = 1 Dxx+Dx /2h(h)dh.x-Dx /2(13)semi-discr

34、ete evolution. A mixed operator shares the char- acters of both fast and slow schemes.Its evident that the scheme (11) shows different dissipa- tion and dispersion properties when s chooses different value. According to the GVC theory, a “fast/mixed type”j +1/21.2 WENO schemes 10,20scheme f(sm) shou

35、ld be used at the left side of dis-In the WENO method, there are r candidate stencils, eachcontinuity and a slow type scheme fj +1/2 (ss ) must be atcontaining r grid points. The one most upwind candidate stencil ranges over mesh point indices j-(r-1) to j, while the fully downwind candidate stencil

36、 ranges over j to j+(r-1), and the other candidate stencils fall in between. The final numerical approximation becomesthe right side of the discontinuity (assume the discontinuity propagates towards right) 18. The two parameters sm andss will be determined in sect. 2.3.Note that the fifth-order line

37、ar upwind scheme (LUW5) and the seventh-order linear upwind scheme (LUW7) aref= r -1w f.(7)gained when s=s0. That is to say we can get a 2r-1 orderj +1/2l =0l j +1/2,lscheme when we choose s=s0. Values of al and bl are listed in Tables 1 and 2, respectively. And values of s0 are listedj +1/2,lSpecif

38、ically,fare rth-order accurate polynomialin Table 3.interplant evaluated at xj+1/2. The weights are defined byr -1 xl 22.2 Weighted methodology for selection of mixed orslow operatorb = j+1/2 Dx2l -1 d f dx,(8)kl l =0 x j-1/2 dx Consider eq. (1), we propose the following weighted meth-odology to con

39、struct WGVC schemes.Firstly, we list the two constraints that WGVC schemes should satisfyIn smooth region, fj +1/2 = fj +1/2 (s0 ) . The two constraints listed above are thus satisfied.j +1/ 2(1) The 2(r-1) order mixed schemef(s m) is used2.3 Determination of sm and ssbehind a discontinuity while th

40、e 2(r-1) order slow schemejf +1/ 2 (s s ) is used ahead of the discontinuity.There are two methods to determined the two parameters sm) and f(2) In the smooth region, the two 2(r-1) order schemesand ss. One method is the optimization method 22. For a nonlinear scheme, no analytical formula of the sp

41、ectral rela-fj +1/ 2(smj +1/ 2 (ss ) are combined to a scheme oftions can be obtained. However, using the technique sug-j +1/2 ( 0 )m j +1/2 ( m )2r-1order,thatis,fs= D fsgested by Pirozzoli 23, the approximate dispersion relation(ADR) of the nonlinear scheme still can be obtained. The+D fs j +1/2(s

42、 s) . Values of the optimal weights Dm and Dstwo parameters sm and ss can be determined numerically.are listed in Table 3.Secondly, as shown in Figure 1, the whole stencil of the scheme (12) is divided into two disjointed sub-stencils de- noted by Sm and Ss.Calculate the WENO smooth indicators of th

43、is two sub- stencils bm and bs following eq. (8). Once there is a discontinuity in sub-stencil Ss, this indicates that this point j is behind the discontinuity, thus the mixed schemeThe other method is to optimize the shockwave-capturing capability using the one-dimensional Sod shock tube prob- lem

44、24. Details can be found elsewhere 18. In this paper, we use this method and values of the two parameters sm andss are listed in Table 3.In critical points, the WGVC scheme reduce to order of 2(r-1) at most while the WENO scheme reduce to order ofr. And the actual ADRs of these nonlinear schemes are

45、fj +1/ 2(sm) should be used. Similarly, once a discontinuityshown in sect. 4.1. These results show that WGVC schemeshave high-resolution for short waves. In additional, sincejis in the sub-stencil Sm, it means this point is ahead of thethe WGVC schemes belong to the GVC schemes. they share discontin

46、uity, thus the slow schemef +1/ 2 (s s ) should bethe moderate shock-capturing capability of the GVCused. And the weights of the two sub-stencils are calculated byscheme 17,18.g k =Dk,(bk + e )p3Hybrid WGVC-WENO schemes(14)In this section, we hybrid the WGVC schemes and the g kWENO schemes of the sa

47、me order to compute multi-scaleswhere k=m or s.v k = g,m + g s(15)flows with discontinuities. The reason are as follows.The common switch functions usually have artificial pa- rameters. It is easy to adjust the parameters so as to identifyThe e is a small positive number which is introduced topreven

48、ts division by zero and we use the classical value ofe=10-6 as in ref. 10. The power parameter p is set to 2 as in ref. 10.Thus we obtain WGVC schemes, that is,less points as discontinuity points. In this situation, the WGVC scheme is used in smooth region (containing some discontinuity points). Thu

49、s the problem the switch function may mistake some points in smooth waves as discontinuitypoints, particularly in high-wavenumber waves 15 can bef= v f(s ) +v f(s ).(16)avoided.j +1/2m j +1/2ms j +1/2sAny switch function can be used. In this paper we useOnce there is a discontinuity in sub-stencil S

50、 , the com-q(s)=sq( q+1-qs) as a switch function, wheres = 1 -sv vD Dputed weightsv m 1,v s 0, thusfj +1/2 = fj +1/2 m s m sand q is a controllable positive integer. Obviously(sm ). Once a discontinuity is in the sub-stencil Sm, the com- the function q remains shares: in smooth regions, q0; and pute

51、d weights v m 0 , v s 1 , thenfj +1/2 = fj +1/2 (s s ) .in discontinuities, q1. Extensive numerical tests show thebest compromise between accuracy and non-oscillatoryproperties is ensured for q=100.4Numerical testsFigure 1 (Color online) Weighted methodology for GVC.In this section, we discuss appli

52、cations of the WGVC-WENO schemes for some benchmark cases.For 1D and 2D compressible fluid problems, the com- pressible Euler equations are solved in the following man- ner. The Roe approximation is used for the local character- istic decomposition at the cell faces while the local Lax- Friedriches

53、flux 25 vector splitting is used to split the fluxes into a positive and a negative part, that is, F = F + + F - , the new schemes are used in each charac- teristic fields. After transforming back into physical space, the numerical flux is then obtained. The 3rd-order TVD Runge-Kutta scheme 25 is used for time integration. All computations are carried out with a CFL=0.6.For 3D tests, the Navier-Stokes equations are solved di- rectly. And the Steger-Warming splitting method is applied while the eight-order central scheme is used to discrete the viscous terms. The ot

温馨提示

  • 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
  • 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
  • 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
  • 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
  • 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
  • 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
  • 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。

评论

0/150

提交评论