6-3 �_���ȻP�_���V�q

¤@­Ó¯x°} $A$ »P¨ä©_²§­È¡]Singular Value¡^$\sigma$ ¤Î©_²§¦V¶q ¡]Singular Vectors¡^$u$ »P $v$ ¤§¶¡¦s¦b¤U¦CªºÃö«Y¦¡¡G $$ \left\{ \begin{matrix} Av=\sigma u\\ A^Tu=\sigma v \end{matrix} \right. $$ ­Y±N©Ò¦³ªº¦æ¦V¶q $u$ ¨Ã±Æ¦¨¯x°} $U$¡A©Ò¦³ªº¦æ¦V¶q $v$ ¨Ã±Æ¦¨¯x°} $V$¡A«h¤W¦¡¥i¼g¦¨¡G

$$ \left\{ \begin{matrix} AV=U\Sigma\\ A^TU=V\Sigma \end{matrix} \right. $$ ¨ä¤¤ $\Sigma$ ªº¥D¹ï¨¤½u§Y¬O¹ïÀ³ªº $\sigma$ ­È¡A¨ä¾l¤¸¯À¬°¹s¡C

­Y $A$ ªººû«×¬O m¡Ñn¡A«h $U$¡B$£U$¡B$V$ ªººû«×¤À§O¬O m¡Ñm¡Bm¡Ñn ¥H¤Î n¡Ñn¡C ¤@¯ë¦Ó¨¥¡A$U$ ©M $V$ §¡¬O Unitary ¯x°}¡]§Y¯x°}¤ºªº¦æ¦V¶q§¡¨â¨â¬Û¤¬««ª½¡A¥B¦æ¦V¶qªºªø«×§¡¬° 1¡^¡Aº¡¨¬¤U¦C±ø¥ó¡G

$$ \left\{ \begin{matrix} UU^T=I\\ VV^T=I \end{matrix} \right. $$ ¦]¦¹¯x°} $A$ ¥i¼g¦¨

$$A=U\Sigma V^T$$ ¤W¦¡ºÙ¬°©_²§­È¤À¸Ñ¡]Singular Value Decomposition¡^¡C

MATLAB ªº svd «ü¥O¥i¥Î©ó­pºâ¯x°}ªº©_²§­È¤Î©_²§¦V¶q¡A§Ú­Ì¨Ã¥i¥H¦P®ÉÅçÃÒ©_²§­È¤À¸Ñ¡A¦p¤U¡G

Example 1: 06-½u©Ê¥N¼Æ/svd01.mA = [3 5; 4 7; 2 1; 0 3]; [U, S, V] = svd(A) maxDiff = max(max(abs(A-U*S*V')))U = -0.5577 0.0881 -0.6954 0.4447 -0.7714 0.0333 0.2489 -0.5848 -0.1771 0.6471 0.5453 0.5025 -0.2504 -0.7565 0.3965 0.4558 S = 10.4517 0 0 1.9397 0 0 0 0 V = -0.4892 0.8722 -0.8722 -0.4892 maxDiff = 3.5527e-015

­Y $A$ ¬° m¡Ñn ªº¯x°}¥B m >> n¡A«h§Ú­Ì¥i¦b­ì¥ýªº svd «ü¥O¥[¤J¥t¤@­Ó¿é¤J¤Þ¼Æ 0¡A¨Ï¨ä²£¥Íªº $U$ ¤Î $S$ ¯x°}¨ã¦³¸û¤pªººû«×¡A¨Ò¦p¡G

Example 2: 06-½u©Ê¥N¼Æ/svd02.mA = [3 5; 4 7; 2 1; 0 3]; [U, S, V] = svd(A, 0) maxDiff = max(max(abs(A-U*S*V')))U = -0.5577 0.0881 -0.7714 0.0333 -0.1771 0.6471 -0.2504 -0.7565 S = 10.4517 0 0 1.9397 V = -0.4892 0.8722 -0.8722 -0.4892 maxDiff = 3.5527e-015

¤@¯ë¦Ó¨¥¡A­Y $A$ ¬° m¡Ñn¡]m >> n¡^ªº¯x°}¡A«h¦¹ºØ¤è¦¡©Ò²£¥Íªº $U$¡B $S$¡B $V$ ¯x°}¡A¨äºû«×¤À§O¬O m¡Ñn¡Bn¡Ñn ¤Î n¡Ñn¡C

©M svd ¬ÛÃöªº«ü¥O©|¦³ svds ¤Î gsvd¡AŪªÌ¥i¥Ñ½u¤W¤ä´©¬d¨ì¬ÛÃö¸ê°T¡C


MATLABµ{¦¡³]­p¡G¶i¶¥½g