| 查看: 3011 | 回复: 15 | ||
| 当前只显示满足指定条件的回帖，点击这里查看本话题的所有回帖 | ||
[求助]
Fortran 求解复数矩阵SVD时，用到cgesvd，结果与matlab不同，为什么？
|
||
|
RT，很小的一个测试矩阵，代码如下 integer M,N,MN,Mmax, Nmax,p, nu, nv parameter (M=6,N=4,MN=min(M,N)) complex :: A(M,N),AA(M,N) character , parameter :: job = 'A' real S(MN) complex :: U(M,M), VT(N,N), V(N,N) complex :: work(1000) real :: rwork(1000) integer info,rw,cl,lwork,kk,i,j,k do rw=1,M do cl=1,N A(rw,cl)=cmplx(rw,2*cl) enddo enddo lwork=MAX(1,2*MIN(M,N)+MAX(M,N)) call CGESVD( JOB, JOB, M, N, A, M, S, U, M, VT, N, & & WORK, LWORK, RWORK, INFO ) V=transpose(conjg(VT)) aa=u*s*v' 得到的u,v只有前2 列与matlab相同，而后面几列的差距很大，S的结果两边差异非常小，基本一致。 而矩阵aa与原始的一样。有大神可告知原因么？ [ Last edited by 长路漫漫 on 2013-6-26 at 11:09 ] |
回复此楼» 猜你喜欢
22408 312求调剂
已经有16人回复
-
通信工程求调剂！！！
已经有3人回复
-
材料299专硕求调剂
已经有16人回复
-
药学求调剂
已经有10人回复
-
材料相关专业344求调剂双非工科学校或课题组
已经有21人回复
-
085404 22408 309分求调剂
已经有7人回复
-
302分求调剂
已经有11人回复
-
271求调剂
已经有29人回复
-
297工科调剂?
已经有13人回复
-
274求调剂
已经有12人回复
» 本主题相关价值贴推荐，对您同样有帮助:
- 【求助】fortran：如何求解 复数矩阵的秩【已完结】
已经有5人回复
9楼2013-06-27 17:52:57
2楼2013-06-26 21:02:44
【答案】应助回帖
★
jjdg: 金币+1, 感谢参与 2013-06-27 00:31:39
jjdg: 金币+1, 感谢参与 2013-06-27 00:31:39
|
具体的讨论，见LAPACK官方论坛： http://icl.cs.utk.edu/lapack-for ... ?f=2&t=1689 当然还有其他地方，都说过类似的问题的 |
3楼2013-06-26 21:03:55
|
谢谢，非常有帮助，只是，我的矩阵很小，而且对精度要求也没有那么高，给你看看一个4×3的复数矩阵得到的两种结果 矩阵A M=4; N=3; for ii=1:M for jj=1:N a(ii,jj)=ii+i*2*jj; end end matlab： u = Columns 1 through 4 -0.2028 - 0.2666i 0.5235 + 0.3708i 0.4803 + 0.0812i 0.3898 - 0.1821i -0.2559 - 0.2442i 0.3188 + 0.2616i -0.4234 - 0.1552i -0.3399 + 0.2662i -0.3091 - 0.2217i 0.1142 + 0.1524i -0.3466 + 0.1235i 0.0841 - 0.0690i -0.3622 - 0.1993i -0.0905 + 0.0432i 0.3628 - 0.0013i -0.5831 + 0.1887i -0.4153 - 0.1768i -0.2952 - 0.0660i -0.3939 - 0.1528i 0.3245 - 0.3249i -0.4684 - 0.1544i -0.4998 - 0.1753i 0.3208 + 0.1047i 0.1246 + 0.1211i Columns 5 through 6 0.0100 - 0.1874i -0.0737 + 0.1148i 0.0538 - 0.3784i 0.2691 - 0.3108i -0.1977 + 0.7171i 0.2468 + 0.2400i 0.0048 + 0.2312i -0.5080 - 0.0163i 0.3183 - 0.1634i -0.4324 - 0.0181i -0.1892 - 0.2192i 0.4982 - 0.0096i s = 32.8054 0 0 0 0 2.7938 0 0 0 0 0.0000 0 0 0 0 0.0000 0 0 0 0 0 0 0 0 v = -0.3202 -0.7730 0.2225 0.5005 -0.3972 + 0.1228i -0.3530 - 0.0509i -0.1119 - 0.3241i -0.7494 + 0.1441i -0.4742 + 0.2455i 0.0671 - 0.1017i -0.4436 + 0.6482i -0.0026 - 0.2882i -0.5512 + 0.3683i 0.4871 - 0.1526i 0.3331 - 0.3241i 0.2515 + 0.1441i zegsvd结果： u= (-0.202824375016695,-0.266642178187997) (-0.255949364969120,-0.244187270428356) (-0.309074354921545,-0.221732362668715) (-0.362199344873970,-0.199277454909075) (-0.415324334826394,-0.176822547149434) (-0.468449324778819,-0.154367639389794) (0.523495728897005,0.370844937064488) (0.318833666239141,0.261621268331258) (0.114171603581277,0.152397599598028) (-9.049045907658722E-002,4.317393086479759E-002) (-0.295152521734451,-6.604973786843274E-002) (-0.499814584392315,-0.175273406601663) (-0.340495458990843,0.541393498514851) (0.337828171542821,-0.540655966822066) (0.160806639553680,-0.275424645436289) (4.866929136520519E-002,7.466583397862278E-002) (-0.228593892608515,0.132598643065634) (2.178524913765310E-002,6.742263669924775E-002) (-0.213595095465232,1.770559300328978E-002) (-8.118515338632207E-002,-0.101568035512989) (0.142978695055282,-2.696899503189933E-002) (0.159236069089761,0.159954322471254) (0.645307867526083,7.874251718869874E-002) (-0.652742382819571,-0.127865402118354) (0.101959655905054,7.515513816888129E-002) (0.385377896491782,-6.292538458577177E-002) (-0.389645765970497,-8.452303330092807E-002) (-0.660124183763325,0.118679170354774) (0.437875799945744,-0.107863393308084) (0.124556597391242,6.147750267112817E-002) (-5.714988270508767E-002,4.358750066103587E-002) (0.164737090936639,-0.319067431094656) (-0.215932231409642,0.700403583527349) (9.841764078738405E-002,-0.548295272551258) (7.776246243304019E-002,5.371201559391532E-002) (-6.783508004233359E-002,6.965960386361419E-002) s= 32.8054068874654 2.79379293216862 9.854419884063358E-016 6.434838164926571E-016 v= (-0.320194781107884,0.000000000000000E+000) (-0.397196078916262,0.122773730945919) (-0.474197376724640,0.245547461891838) (-0.551198674533019,0.368321192837758) (-0.772965265811650,0.000000000000000E+000) (-0.352952325959982,-5.085805228873791E-002) (6.706061389168552E-002,-0.101716104577475) (0.487073553743353,-0.152574156866213) (-0.282895611404907,0.000000000000000E+000) (0.402086157056524,-0.348691818157092) (4.451452010167405E-002,0.697383636314183) (-0.163705065753291,-0.348691818157091) (-0.469009672659151,0.000000000000000E+000) (0.610331955712300,0.210322709402046) (0.186365106552852,-0.420645418804091) (-0.327687389606002,0.210322709402045) 很明显可发现s基本一致，而u和s从第三列开始就区别非常大，跟你 具体的讨论，见LAPACK官方论坛： http://icl.cs.utk.edu/lapack-forum/viewtopic.php?f=2&t=1689 讨论的内容还挺不一样的，我之前实验过DGESVD这个，处理实数矩阵时，对于一般精度的矩阵，U,S,V结果基本差不多，误差非常小了，而这个CGESVD和ZESVD处理复数矩阵，误差非常大，就LAPACK官方论坛上，也没有搜索到与CGESVD和ZESVD处理复数矩阵的类似帖子 为这个问题纠结了很久，麻烦大神进一步解答哈，谢谢～～ |
4楼2013-06-27 11:20:53
5