師大
Power and inverse power methods
Tsung-Ming Huang
Department of Mathematics National Taiwan Normal University, Taiwan
February 15, 2011
Outline
1 Power method
2 Inverse power method
師大
Definition 1
1 An eigenvalue whose geometric multiplicity is less than its algebraic multiplicity isdefective.
2 The matrix A ∈ Cn×n has acomplete systemof eigenvectors if it has n linearly independent eigenvectors.
Let A be a nondefective matrix and (λi, xi)for i = 1, · · · , n be a complete set of eigenpairs of A. That is {x1, · · · , xn} is linearly independent. Hence, for any u0 6= 0, ∃ α1, · · · , αnsuch that
u0 = α1x1+ · · · + αnxn. Now Akxi= λkixi, so that
Aku0= α1λk1x1+ · · · + αnλknxn. (1) If |λ1| > |λi| for i ≥ 2 and α1 6= 0, then
1
λk1Aku0 = α1x1+ (λ2 λ1
)kα2x2+ · · · + αn(λn λ1
)kxn→ α1x1as k → ∞.
Algorithm 1 (Power Method with 2-norm) Choose an initial u 6= 0 with kuk2 = 1.
Iterate until convergence
Computev = Au; k = kvk2; u := v/k Theorem 2
The sequence defined by Algorithm 1 is satisfied
i→∞lim ki = |λ1|
i→∞lim εiui= x1
kx1k α1
|α1|, where ε = |λ1| λ1
師大
Proof: It is obvious that
us = Asu0/kAsu0k, ks= kAsu0k/kAs−1u0k. (2) This follows from λ1−sAsu0 −→ α1x1that
|λ1|−skAsu0k −→ |α1|kx1k
|λ1|−s+1kAs−1u0k −→ |α1|kx1k and then
|λ1|−1kAsu0k/kAs−1u0k = |λ1|−1ks−→ 1.
From (1) follows now for s → ∞
εsus = εs Asu0 kAsu0k =
α1x1+Pn i=2αi
λi
λ1
s
xi kα1x1+Pn
i=2αi
λi
λ1
s
xik
→ α1x1
kα1x1k = x1 kx1k
α1
|α1|.
Algorithm 2 (Power Method with Linear Function) Choose an initial u 6= 0.
Iterate until convergence
Computev = Au; k = `(v); u := v/k
where`(v), e.g. e1(v)oren(v), is a linear functional.
Theorem 3
Suppose `(x1) 6= 0and `(vi) 6= 0, i = 1, 2, . . . ,then
i→∞lim ki = λ1 i→∞lim ui = x1
`(x1).
師大
Proof: As above we show that
ui = Aiu0/`(Aiu0), ki= `(Aiu0)/`(Ai−1u0).
From (1) we get for i → ∞
λ1−i`(Aiu0) −→ α1`(x1), λ1−i+1`(Ai−1u0) −→ α1`(x1), thus
λ1−1
ki−→ 1.
Similarly for i −→ ∞,
ui= Aiu0
`(Aiu0) = α1x1+Pn
j=2αj(λλj
1)ixj
`(α1x1+Pn
j=2αj(λλj
1)ixj)
−→ α1x1
α1`(x1)
Note that:
ki = `(Aiu0)
`(Ai−1u0) = λ1
α1`(x1) +Pn
j=2αj(λλj
1)i`(xj) α1`(x1) +Pn
j=2αj(λλj
1)i−1`(xj)
= λ1+ O
| λ2
λ1
|i−1
.
That is the convergent rate is
λ2
λ1
.
師大
Theorem 4
Let u 6= 0 and for any µ set rµ= Au − µu. Then krµk2is minimized when
µ = u∗Au/u∗u.
In this case rµ⊥ u.
Proof: W.L.O.G. assume kuk2= 1. Let u U be unitary and set
u∗ U∗
A u U ≡
ν h∗ g B
=
u∗Au u∗AU U∗Au U∗AU
.
Then
u∗ U∗
rµ =
u∗ U∗
Au − µ
u∗ U∗
u
=
u∗ U∗
A u U
u∗ U∗
u − µ
u∗ U∗
u
=
ν h∗
g B
u∗ U∗
u − µ
u∗ U∗
u
=
ν h∗
g B
1 0
− µ
1 0
=
ν − µ g
.
It follows that krµk22= k
u∗ U∗
rµk22 = k
ν − µ g
k22 = |ν − µ|2+ kgk22.
師大
Hence
minµ krµk2 = kgk2 = krνk2. That is µ = ν = u∗Au. On the other hand, since
u∗rµ= u∗(Au − µu) = u∗Au − µ = 0, it implies that rµ⊥ u.
Definition 5 (Rayleigh quotient)
Let u and v be vectors with v∗u 6= 0.Then v∗Au/v∗uis called a Rayleigh quotient.
If u or v is an eigenvector corresponding to an eigenvalue λ of A, then v∗Au
v∗u = λv∗u v∗u = λ.
Therefore, u∗kAuk/u∗kukprovide a sequence of approximation to λ in the power method.
Inverse power method
Goal
Find the eigenpair (λ, x) of A where λ is belonged to a given region or closest to a certain scalar σ.
Let λ1, · · · , λnbe the eigenvalues of A. Suppose λ1 is simple and σ ≈ λ1.Then
µ1 = 1
λ1− σ, µ2 = 1
λ2− σ, · · · , µn= 1 λn− σ
are eigenvalues of (A − σI)−1 and µ1→ ∞ as σ → λ1. Thus we transform λ1 into a dominant eigenvalue µ1.
The inverse power method is simply the power method applied to (A − σI)−1.
師大
Algorithm 3 (Inverse power method with a fixed shift) Choose an initial u0 6= 0.
For i = 0, 1, 2, . . .
Computevi+1= (A − σI)−1ui andki+1= `(vi+1).
Setui+1= vi+1/ki+1
The convergence of Algorithm 3 is |λλ1−σ
2−σ| whenever λ1 and λ2 are the closest and the second closest eigenvalues to σ.
Algorithm 3 is linearly convergent.
Let (λ, x) be an eigenpair of A, i.e.,
Ax = λx ⇒ (A − σI)x = (λ − σ)x ⇒ (A − σI)−1x = 1
λ − σx ≡ µx.
It implies that
λ = σ + µ−1.
Algorithm 4 (Inverse power method with variant shifts) Choose an initial u0 6= 0. Given σ0= σ.
For i = 0, 1, 2, . . .
Computevi+1= (A − σiI)−1ui andki+1= `(vi+1).
Setui+1= vi+1/ki+1andσi+1= σi+ 1/ki+1.
師大
Connection with Newton method
Consider the nonlinear equations:
F
u λ
≡
Au − λu
`Tu − 1
=
0 0
. (3)
Newton method for (3): for i = 0, 1, 2, . . .
ui+1 λi+1
=
ui λi
−
F0
ui λi
−1
F
ui λi
. Since
F0
u λ
=
A − λI −u
`T 0
, the Newton method can be rewritten by component-wise
(A − λiI)ui+1 = (λi+1− λi)ui (4)
`Tui+1 = 1. (5)
Let
vi+1= ui+1 λi+1− λi. Substituting vi+1into (4), we get
(A − λiI)vi+1= ui. By equation (5), we have
ki+1= `(vi+1) = `(ui+1)
λi+1− λi = 1 λi+1− λi. It follows that
λi+1= λi+ 1 ki+1.
師大
Algorithm 5 (Inverse power method with Rayleigh Quotient) Choose an initial u0 6= 0 with ku0k2 = 1.
Compute σ0= uT0Au0. For i = 0, 1, 2, . . .
Computevi+1= (A − σiI)−1ui.
Setui+1= vi+1/kvi+1k2andσi+1= uTi+1Aui+1. For symmetric A, Algorithm 5 is cubically convergent.