Professional Documents
Culture Documents
Eval Norms
Eval Norms
3. Symmetric eigendecomposition
• symmetric eigendecomposition
• quadratic forms
3.1
Eigenvalues and eigenvectors
Ax = λx
• this immediately shows that every square matrix has at least one eigenvalue
• the roots of the polynomial (and corresponding eigenvectors) may be complex
• (algebraic) multiplicity of an eigenvalue is its multiplicity as a root of det(λI − A)
• there are exactly n eigenvalues, counted with their multiplicity
• set of eigenvalues of A is called the spectrum of A
A11
0 ··· 0
0 A22 · · · 0
A = .
.. ... ..
.
0
0 ··· Ann
Aei = Aii ei
B = T −1 AT
By = (T −1 AT)(T −1 x) = T −1 Ax = T −1(λx) = λy
T −1 AT = Λ
λ1 0 · · · 0 w1T
0 λ2 · · ·
−1 0 w2T
A = TΛT =
v1 v2 · · · vn .. .. . . . ..
..
λn
0 0 ···
wnT
= λ1 v1 w1T + λ2 v2 w2T + · · · + λn vn wnT
x = TT −1 x = α1 v1 + · · · + αn vn where αi = wiT x
Ax = α1 λ1 v1 + · · · + αn λn vn = TΛT −1 x
a1 an an−1 ··· a3 a2
a2 a1 an ··· a4 a3
a3 a2 a1 ··· a5 a4
A =
.. .. .. ... .. ..
an−1 an−2 an−3 ··· a1 an
an an−1 an−2 ··· a2 a1
1
W −1 = W H
n
• symmetric eigendecomposition
• quadratic forms
QT AQ = Λ, QT Q = I
Ax = λx, x,0
n X
n n
x Ax =
H
Ai j x̄i x j = Aii |xi | + 2
2
X X X
Ai j Re( x̄i x j )
i=1 j=1 i=1 j<i
λyT x = yT Ax = xT Ay = µxT y
A = QΛQT (1)
• Q is orthogonal
λ1 0 · · · 0
0 λ2 · · ·
0
=
A q1 q2 · · · qn q1 q2 · · · qn . .. . . . ..
.
λn
0 0 ···
= λ1 q1 λ2 q2 · · · λn qn
q1T λ1 λ1 0
0
QT AQ = A q1 V Q̃ = =
Q̃T V T 0 Q̃T V T AV Q̃ 0 Λ̃
Symmetric eigendecomposition 3.12
Spectral decomposition
λ1 0 · · · 0 q1T
0 λ2 · · ·
0 q2T
A = QΛQ T
=
q1 q2 · · · qn .. .. . . . ..
..
λn
0 0 ···
qnT
n
= λi qi qiT
X
i=1
n
Ax = λi qi (qiT x)
X
i=1
λ1 · · · 0 q1T
A = QΛQT =
q1 · · ·
qn .. . . . ..
..
λn
0 ···
qnT
Ordering of eigenvalues
diagonal Λ and columns of Q can be permuted; we will assume that
λ1 ≥ λ2 ≥ · · · ≥ λn
Choice of eigenvectors
suppose λi is an eigenvalue with multiplicity k : λi = λi+1 = · · · = λi+k−1
a symmetric matrix is invertible if and only if all its eigenvalues are nonzero:
• inverse of A = QΛQT is
1/λ1
0 ··· 0
0 1/λ2 · · · 0
A−1 = (QΛQT )−1 = QΛ−1QT , Λ−1 = .
. .. ... ..
0
0 ··· 1/λn
Integer powers
Square root
p p
A 1/2
= QΛ Q ,
1/2 T
Λ1/2
= diag( λ1, . . . , λn)
Other matrix functions: can be defined via power series, for example,
QT1
Λ1 0
A = QΛQT = = Q1Λ1QT1
Q1 Q2
0 0 QT2
B = Q1, C = Λ1QT1
QT1
Λ1 0
A= = Q1Λ1QT1
Q1 Q2
0 0 QT2
Λ−1 QT1
0
A† = Q1Λ−1 1 =
T
Q Q1 Q2 1
1 0 0 QT2
n
trace(B) =
X
Bii
i=1
n X
m
trace(BC) = trace(CB) =
X
Bi j C ji
i=1 j=1
n
trace(QΛQ ) = trace(Q QΛ) = trace(Λ) =
T T
λi
X
i=1
kUBV kF = kBkF
Proof:
kBxk
kBk2 = max
x,0 k xk
kUBV k2 = kBk2
Proof:
Exercise 1
A − αI
Exercise 2
A = UU T (where U T U = I )
Exercise 3
• symmetric eigendecomposition
• quadratic forms
the eigendecomposition is a useful tool for problems that involve quadratic forms
f (x) = xT Ax
maximize xT Ax minimize xT Ax
subject to xT x = 1 subject to xT x = 1
Exercise 1: find the extreme values of the Rayleigh quotient (xT Ax)/(xT x), i.e.,
xT Ax xT Ax
max T , min T
x,0 x x x,0 x x
maximize xT Ax minimize xT Ax
subject to xT x ≤ 1 subject to xT x ≤ 1
xT Ax xT Ax
λ1 = max x Ax = max T ,
T
λn = min x Ax = min T
T
k xk=1 x,0 x x k xk=1 x,0 x x
E = {x | xT Ax ≤ 1}
√1 q1
λ1
√1 qn
λn
λ1 y12 + · · · + λn yn2 ≤ 1
E = {x | xT Ax ≤ 1}
X=
q1 q2 · · · qk
maximize λmin(Y T ΛY )
subject to Y TY = I
we also partition Λ as
λ1 · · · 0 λ k+1 · · · 0
Λ=
Λ1 0
, Λ1 =
.. . . . .. ,
.
Λ2 = . ... ..
0 Λ2
λk λn
0 ···
0
···
I
we show that the matrix Ŷ = is optimal
0
• for this matrix
T
I Λ1 0 I
Ŷ T ΛŶ = = Λ1, λmin(Ŷ T ΛŶ ) = λmin(Λ1) = λ k
0 0 Λ2 0
λmin(Y T ΛY ) = min uT (Y T ΛY )u
kuk=1
λmin(Y T ΛY ) = min uT (Y T ΛY )u ≤ vT (Y T ΛY )v ≤ λ k
kuk=1
Symmetric eigendecomposition 3.31
Min–max characterization of eigenvalues
X=
qn−k+1 · · · qn−1 qn
A11
A12 · · · A1m
A21 A22 · · · A2m
B = . ,
.. .. (4)
.
Am1
Am2 · · · Amm
show that
µ1 ≤ λ1, µ2 ≤ λ2, . . ., µm ≤ λm
λ1 ≥ µ1 ≥ λ2 ≥ µ2 ≥ · · · ≥ λn−1 ≥ µn−1 ≥ λn
x2
contour lines of density function for
√
1
Σ= √7 3
, µ=
5
4 3 5 4
eigenvalues of Σ are λ1 = 2, λ2 = 1,
√
3/2 1/2
q1 = , q2 = √
1/2 − 3/2
x1
n
1 yi2
p̃(y) =
Y
√ exp(− )
i=1 2πλi 2λi
x2
y2
y1
y1
−λ11/2 λ11/2
y2 x1
−λ21/2 λ21/2
QT AQ = Λ
ST AS = D, ST BS = I
maximize xT Ax
subject to xT Bx = 1
ST AS = D, ST BS = I
maximize yT Dy
subject to yT y = 1
• symmetric eigendecomposition
• quadratic forms
• low rank is not a robust property (easily destroyed by noise or estimation error)
the best rank-r approximation is the sum of the first r terms in the decomposition:
r
B= λi qi qiT
X
i=1
• B is the best approximation for the Frobenius norm: for every C with rank r ,
! 1/2
n
k A − CkF ≥ k A − BkF = λi2
X
i=r+1
• B is also the best approximation for the 2-norm: for every C with rank r ,
• the first order optimality conditions will be derived on page 3.41; they are
4(A − X X T )X = 0
Symmetric eigendecomposition 3.41
Solution of first order optimality conditions
AX = X(X T X)
AY = Y D, Y TY = D
we conclude that the solutions of the first order optimality conditions satisfy
X X = YY =
T T
λi qi qiT
X
i∈I
among the solutions of the 1st order conditions we choose the one that minimizes
k A − X X T kF
= λi qi qiT kF2
X
k
i<I
= λi2
X
i<I
minimize k A − X X T kF2
4(A − X X T )X = 0
the same matrix B is also the best approximation in 2-norm: if C has rank r , then
k A − Ck2 ≥ k A − Bk2
n r
k A − Bk2 = λi qi qiT λi qi qiT k2
X X
k −
i=1 i=1
n
= λi qi qiT k2
X
k
i=r+1
= λr+1