ECE735 FUNDAMENTALS OF NETWORK SCIENCE
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
ECE735
FUNDAMENTALS OF NETWORK SCIENCE
Matrices: Essential Notions
HW 23 0906
Department of Electrical and Computer Engineering
Question |
Points |
Question 1 |
20 |
Question 2 |
20 |
Question 3 |
20 |
Question 4 |
20 |
Question 5 |
20 |
I. QUESTION
Assuming the matrix inverses appearing in the expressions exist, prove the following relation- ships involving the matrix inverse and matrix determinant:
1) Block inversion formula:
A = »A1 A2 fi
–A3 A4 fl
=》 A´1 = » `A1 - A2 A4(´)1 A3 ˘´1 -A1(´)1 A2 `A4 - A3 A1(´)1 A2 ˘´1fi
– -A4(´)1 A3 `A1 - A2 A4(´)1 A3 ˘´1 `A4 - A3 A1(´)1 A2 ˘´1 fl.
2) For matrices A, B , C, and D (of appropriate size),
`A + B D ´1C ˘´1 = A´1 - A´1B `D + CA´1B ˘´1 CA´1 .
3) For matrices A P CMXN and B P CNXM ,
pIM ` A Bq — 1 A“ A pIN ` B Aq — 1 .
4) For matrices A P CMXN and B P CNXM ,
det pIM ` A Bq“ det pIN ` AT BT q.
II. QUESTION
Consider the column vectors u P CM and v P CN . Suppose E “ u vH . Show the following:
1) }E }F “ }E }2 “ }u }2 }v }2 .
2) }E }构 “ }u }构 ¨ }v }1 and }E }1 “ }u }1 ¨ }v }构 .
III. QUESTION
Consider the SVD of the matrix A P CMXN :
n
A“ V Σ UH “ÿ σiviui(H), n“ mintM, Nu,
i=1
where ui and vi are the columns of the unitary matrices U and V , respectively; Σ is the ‘diagonal’ matrix with the singular values σi of A populating its diagonal. Consider the linear mapping
Ax, where x P CN :
Stage III(ÝÝÝÝÝÑ)
We may view the linear mapping Ax as being composed of the following stages: . Stage I Rotation by UH. Decompose x into the ‘coordinates’
»— u1(H)xfiffi
UH x“ — ffi .
–uN(H)xfl
Geometrically, this corresponds to a rotation of x by UH .
. Stage II Dilation by Σ. Scale the coordinate ui(H)x by σi :
σ(σ)n(1)x(x) ffi “ Σ UH x, where n“ mintM, Nu.
–0(M —n)X1fl
Geometrically, this corresponds to a dilation of UH x by Σ .
. Stage III Rotation by V. Reconstitute Ax by associating the scaled coordinate σiui(H)x with the direction of vi :
l σ 1 u1(H)x 」
' .(.) ' n n
[v 1 . . . vn vn+1 . . . vM ] ' σn u(.)n(H)x ' = (σiui(H)x) vi = σiviui(H) = Ax.
「0(M —n)X1l
Geometrically, this corresponds to a rotation of Σ UH x by V.
1) Use the matrix
A = l — 1\2 3\2 」
「 3\2 — 1\2l
to illustrate this decomposition. Use the 2-D plane to sketch what occurs at each stage. 2) When A e CNXN is Hermitian symmetric, its SED (symmetric eigenvalue decomposition)
yields
A = V Λ VH = λivivi(H) .
Here, as usual,
Λ = diag {λi ,...,λN }, and V = [v 1 . . . v N ] ,
are the diagonal matrix of the eigenvalues of A and the unitary matrix formed from the eigenvectors of A, respectively. So, the SED can also be given a similar geometric interpretation as that for the SVD. What is the main difference between these interpretations
given for the SVD and the SED?
IV. QUESTION
An ellipsoid in CN centered at x0 e CN is an affine transformation of the unit ball (in terms of the 2-norm) in CN , i.e,
{xe CN | x = x0 + B z , where IzI2 会 1 and B e CN XN is non-singular} .
1) For a given A > 0 (i.e., A is p.d.) , show that the set
{xe CN | (x — x0 )HA (x — x0 ) 会 1}
is an ellipsoid.
2) Show that the semiaxes of this ellipsoid lie along the eigenvectors v 1 and vN , which correspond to the eigenvectors λ1 = λmax and λN = λmin , respectively.
3) What is the ‘thickness’ of the ellipsoid along each semiaxis?
4) Plot the ellipsoid corresponding to
A = »3 1fi ,
– 1 4fl
and identify its semiaxes and their directions, and the ‘thickness’ along each semiaxis direction.
V. QUESTION
Consider the matrix A E CMˆN with singular values σi , i E 1,n, n = min{M, N} and rank (A) = r < n. Find a matrix B E CNˆN s.t.
= AB
would have the same left and right singular vectors as A, and its singular values are given by
i = &’% σi , for i E m + 1,n,
he(.e.), h(h)e(a)s(s)ttsin(he)g(s)u(a)la(m)r(e) v(s)al(in)u(g)fvA(al)ues as A, except that its highest m singular values are αi times
2023-09-05
Matrices: Essential Notions