第34卷 第6期 上海师范大学学报(自然科学版) Vol.34,No.6
2005 年12 月 JournalofShanghaiNormalUniversity(NaturalSciences) 2005,Dec.
O n an iterative m ethod for solving
the rank deficient linear least squares problem
SU N Le-ping
( M athem atics and Sciences C ollege, Shanghai N orm al U niversity, Shanghai200234, China)
A bstract:
W e apply an iterative m ethod for finding the least squares solution to the inconsistent system A x = b ,
whereAisam ×nmatrixofrankr.Themethodisan iteration schem e for consistent system oflinear equations ^Az =
^b
w h ich is a sso ciated w ith
Ax = b
. It denotes that under som e conditions, there exists a real num b er
ρ
such that
w heneverγis a com plex num ber with0 <
γ
< ρ, the sequence x
0
,x
1
,x
2
, converges to th e least sq uares solution
of the system A x = b for every initialvectorx
0
,where
^A+γI
()
x
i
=γx
i- 1
+^b,fori = 1,2,….Sometim e s th is
m e th o d is m o re s im p le th an th ose a lready reported.
K ey w ord s:
least squares solution; iteration m ethod; consistent linear system ; Jordan canonical form ; eigenvalue
CLC number:
O 151.21
Documentcode:
A
A r tic le ID :
1000-5137(2005)06-0006-05
R eceived date: 2005-09-08
Biography: SUN Le-ping( 1963-), fem ale, associate professor, M athem atics and Sciences C ollege, ShanghaiNormalUni-
versity.
1 In tro d u c tion
C onsider the problem of obtaining the least squares solution to the system
Ax = b, (1)
w here A is a com plex m ×n m atrix of rank rand b is a com plex m - vector. Fam ous iterative m ethods know n as
successive overrelaxation m ethods( SO R ) and accelerate overrelaxation m ethods( A O R ) have been suggested
by m any people [3,4,6,7 ]. They developed these m ethods, w hich split the aug m e n te d c o e ffic ie n t m a trix b y
SOR orAOR subproper splittin g a n d d e te rm in e th e in te rv a ls fo r th e re la x a tio n p a ra m e te rs w h e re the SO R or
A O R ite ra tio n m a trix is se m ic o n v e rg e n t. H o w e v e r , th e ite ra tio n m a trix H
ω
1
,ω
2
, w hich is very im portant in SO R
o r A O R ite ra tio n z
i+1
=:H
ω
1
,ω
2
z
i
+ c , appears to be com plicated and som ew hat difficult to com pute.
In this paper, w e develop an iterative m ethod
^A+γI
()
x
i
=γx
i- 1
+^b, i = 1,2,…, (* )
fo r e v e ry in itia l v e c to r x
0
, w here γis a com plex num ber w ith 0 <
γ
<ρ,ρis som e real num ber. It can be
p roved that th e sequ en ce x
0
,x
1
,x
2
, generated from the iteration (* ) converges to the least squares solution of
the system Ax = b . A nd this m ethod is som etim es m ore convenient and easier to com pute.
In the follow ing prelim inaries, w e introduce system ^Az = ^b , w h ic h is a s so c ia te d w ith A x = b , a n d so m e
u se fu l re s u lts . O u r m a in re s u lts a re p re s e n te d in se c tio n 3 .