This preview shows page 1. Sign up to view the full content.
Unformatted text preview: hat diagonal matrix AT A are the columns of I , so
V D I in the SVD. Then the ui are Avi =i which is the unit vector wi =i . The SVD of this A with orthogonal columns is A D U †V T D .A† 1 /.†/.I /: 2
2
12 Since AT D A we have 1 D 2 and 2 D 2 . But 2 is negative, so 1 D 3 and
1
2 13
14
15
16 17 2 D 2. The unit eigenvectors of A are the same u1 D v1 as for AT A D AAT and
u2 D v2 (notice the sign change because 2 D 2 , as in Problem 4).
Suppose the SVD of R is R D U †V T . Then multiply by Q to get A D QR. So the
SVD of this A is .QU /†V T . (Orthogonal Q times orthogonal U D orthogonal QU .)
The smallest change in A is to set its smallest singular value 2 to zero. See # 7.
The singular values of A C I are not j C 1. They come from eigenvalues of
.A C I /T .A C I /.
This simulates the random walk used by Google on billions of sites to solve Ap D p.
It is like the power method of Section 9:3 except that it follows the links in one “walk”
where the vector pk D Ak p0 averages over all walks.
p
p
A D U †V T D Œcosines including u4 diag.sqrt.2
2; 2; 2 C 2// Œsine matrixT .
AV D U † says that differences of sines in V are cosines in U times ’s.
The SVD of the derivative on Œ0; with f .0/ D 0 has u D sin nx , D n, v D cos nx ! Problem Set 7.1, page 380
1 With w D 0 linearity gives T .v C 0/ D T .v/ C T .0/. Thus T .0/ D 0. With c D 1
linearity gives T . 0/ D T .0/. This is a second proof that T .0/ D 0.
2 Combining T .c v/ D cT .v/ and T .d w/ D d T .w/ with addition gives T .c v C d w/ D
cT .v/ C d T .w/. Then one more addition gives cT .v/ C d T .w/ C eT .u/.
3 (d) is not linear. Solutions to Exercises 76
4 (a) S.T .v// D v (b) S.T .v1 / C T .v2 // D S.T .v 1 // C S.T .v2 //. 5 Choose v D .1; 1/ and w D . 1; 0/. Then T .v/ C T .w/ D .v C w/ but T .v C w/ D .0; 0/. 6 (a) T .v/ D v=kvk does not satisfy T .v C w/ D T .v/ C T .w/ or T .c v/ D cT .v/ (d) satisﬁes T .c v/ D cT .v/. (b) and (c) are linear 7 (a) T .T .v // D v (b) T .T .v// D v C .2; 2/ (c) T .T .v // D v (d) T .T .v// D T .v/. 8 (a) The range of T .v1 ; v2 / D .v1 v2 ; 0/ is the line of vectors .c; 0/. The nullspace
is the line of vectors .c; c/.
(b) T .v1 ; v2 ; v3 / D .v1 ; v2 / has Range R2 , kernel
{(0; 0; v3 )}
(c) T .v/ D 0 has Range f0g, kernel R2
(d) T .v1 ; v2 / D .v1 ; v1 /
has Range = multiples of .1; 1/, kernel = multiples of .1; 1/. 9 If T .v1 ; v2 ; v3 / D .v2 ; v3 ; v1 / then T .T .v// D .v3 ; v1 ; v2 /; T 3 .v/ D v; T 100 .v/ D T .v /. 10 (a) T .1; 0/ D 0 (b) .0; 0; 1/ is not in the range
n m (c) T .0; 1/ D 0. 11 For multiplication T .v/ D Av: V D R , W D R ; the outputs ﬁll the column space; v is in the kernel if Av D 0. 12 T .v/ D .4; 4/I .2; 2/I .2; 2/; if v D .a; b/ D b.1; 1/C a 2 b .2; 0/ then T .v/ D b.2; 2/C .0; 0/. 13 The distributive law (page 69) gives A.M1 C M2 / D AM1 C AM2 . The distributive law over c ’s gives A.cM / D c.AM /. 14 This A is invertible. Multiply AM D 0 and AM D B by A 1 to get M D 0 and
B . The kernel contains only the zero matrix M D 0....
View
Full
Document
This note was uploaded on 09/25/2012 for the course PHY 103 taught by Professor Minki during the Spring '12 term at Korea Advanced Institute of Science and Technology.
 Spring '12
 Minki
 Mass

Click to edit the document details