Introduction to Numerical Analysis

(Eigenvalues and Eigenvectors: Power Methods)

MA 214, Spring 2023-24.

MA 214 - NA Spring 2023-24 1 / 46

——————————————————————————

MA 214 - NA Spring 2023-24 2 / 46

Eigenvalue of a Matrix

Let A be an n × n matrix with real entries. Eigenvalues of A are defined

as the roots of the equation

det(λI − A) = 0.

MA 214 - NA Spring 2023-24 2 / 46

Eigenvalue of a Matrix

Let A be an n × n matrix with real entries. Eigenvalues of A are defined

as the roots of the equation

det(λI − A) = 0.

Note that det(λI − A) is a polynomial in λ of degree n, which is known

as characteristic polynomial of the matrix A.

MA 214 - NA Spring 2023-24 2 / 46

Eigenvalue of a Matrix (contd.)

When n = 2 the characteristic polynomial is a quadratic polynomial for

which there is a nice formula for computing roots.

MA 214 - NA Spring 2023-24 3 / 46

Eigenvalue of a Matrix (contd.)

When n = 2 the characteristic polynomial is a quadratic polynomial for

which there is a nice formula for computing roots.
When n = 3, there is a formula that many of us do not remember.

MA 214 - NA Spring 2023-24 3 / 46

Eigenvalue of a Matrix (contd.)

When n = 2 the characteristic polynomial is a quadratic polynomial for

which there is a nice formula for computing roots.
When n = 3, there is a formula that many of us do not remember.
When n = 4, there is a formula. But its too complicated.

MA 214 - NA Spring 2023-24 3 / 46

Eigenvalue of a Matrix (contd.)

When n = 2 the characteristic polynomial is a quadratic polynomial for

which there is a nice formula for computing roots.
When n = 3, there is a formula that many of us do not remember.
When n = 4, there is a formula. But its too complicated.
Therefore, numerically approximating the eigenvalues is the only way
out.

MA 214 - NA Spring 2023-24 3 / 46

Eigenvalue of a Matrix (contd.)

One obvious way of approximating an eigenvalue of a matrix is

MA 214 - NA Spring 2023-24 4 / 46

Eigenvalue of a Matrix (contd.)

One obvious way of approximating an eigenvalue of a matrix is

to first obtain the polynomial

det(λI − A) = 0.

explicitly in λ and

MA 214 - NA Spring 2023-24 4 / 46

Eigenvalue of a Matrix (contd.)

One obvious way of approximating an eigenvalue of a matrix is

to first obtain the polynomial

det(λI − A) = 0.

explicitly in λ and
then use some nonlinear solver to compute a root of this
polynomial.

MA 214 - NA Spring 2023-24 4 / 46

Eigenvalue of a Matrix (contd.)

One obvious way of approximating an eigenvalue of a matrix is

to first obtain the polynomial

det(λI − A) = 0.

explicitly in λ and
then use some nonlinear solver to compute a root of this
polynomial.
But this is not an efficient way of computing eigenvalues because of two
reasons.
MA 214 - NA Spring 2023-24 4 / 46
Eigenvalue of a Matrix (contd.)

One reason is that obtaining explicit form of the polynomial is itself a

difficult task when the dimension of the matrix is very large.

MA 214 - NA Spring 2023-24 5 / 46

Eigenvalue of a Matrix (contd.)

One reason is that obtaining explicit form of the polynomial is itself a

difficult task when the dimension of the matrix is very large.

If however one does manage to approximate the characteristic

polynomial by another monic polynomial, then the roots of this
approximated polynomial are close to the eigen values.

MA 214 - NA Spring 2023-24 5 / 46

Eigenvalue of a Matrix (contd.)

Without monicity, roots of two very ‘close’ polynomials may be far

apart.

MA 214 - NA Spring 2023-24 6 / 46

Eigenvalue of a Matrix (contd.)

Without monicity, roots of two very ‘close’ polynomials may be far

apart.
For instance the roots of the polynomials X − 1 and 0.01X2 + X − 1 are
far apart.

MA 214 - NA Spring 2023-24 6 / 46

———————————————————————————-

MA 214 - NA Spring 2023-24 7 / 46

Eigenvalue Problem: Power Method

Power method is a technique to obtain a specific eigenvalue called the

dominant eigenvalue and the corresponding eigen vector for a given
n × n matrix A.

MA 214 - NA Spring 2023-24 7 / 46

Eigenvalue Problem: Power Method

Power method is a technique to obtain a specific eigenvalue called the

dominant eigenvalue and the corresponding eigen vector for a given
n × n matrix A.
Definition (Dominant Eigenvalue of a Matrix)
Let A be an n × n matrix and let µ1 , µ2 , · · · , µn be the eigenvalues of A
(repeated according to their algebraic multiplicities). These eigenvalues
can be re-named as λ1 , λ2 , · · · , λn such that they satisfy the condition:

|λ1 | ≥ |λ2 | ≥ · · · ≥ |λn |.

Then λ1 is said to be the dominant eigenvalue of A.

MA 214 - NA Spring 2023-24 7 / 46
Recall

This notion was useful in the previous lecture too :

Let us consider an iterative method in the form

x(k+1) = Bx(k) + c, k = 0, 1, 2, · · · . (∗)

MA 214 - NA Spring 2023-24 8 / 46

Recall

This notion was useful in the previous lecture too :

Let us consider an iterative method in the form

x(k+1) = Bx(k) + c, k = 0, 1, 2, · · · . (∗)

Theorem (Necessary and Sufficient Conditions)

For any x(0) ∈ Rn , the sequence {x(k) } defined by (∗) converges to the
solution of x = Bx + c if and only if ρ(B) < 1
Here ρ(B) was the spectral radius of B, defined as the modulus of the
dominant eigenvalue.
MA 214 - NA Spring 2023-24 8 / 46
Eigenvalue Problem: Power Method (contd.)

Example:
A given matrix may have unique or more than one dominant (but
distinct or repeated) eigenvalues as illustrated here.

MA 214 - NA Spring 2023-24 9 / 46

Eigenvalue Problem: Power Method (contd.)

Example:
A given matrix may have unique or more than one dominant (but
distinct or repeated) eigenvalues as illustrated here.
The matrix  
1 0 0
A= 0 −2 1
0 0 −1

MA 214 - NA Spring 2023-24 9 / 46

Eigenvalue Problem: Power Method (contd.)

Example:
A given matrix may have unique or more than one dominant (but
distinct or repeated) eigenvalues as illustrated here.
The matrix  
1 0 0
A= 0 −2 1
0 0 −1
has eigenvalues 1, −1, and −2.

MA 214 - NA Spring 2023-24 9 / 46

Eigenvalue Problem: Power Method (contd.)

Example:
A given matrix may have unique or more than one dominant (but
distinct or repeated) eigenvalues as illustrated here.
The matrix  
1 0 0
A= 0 −2 1
0 0 −1
has eigenvalues 1, −1, and −2.
Take λ1 = −2, λ2 = −1, and λ3 = 1.

MA 214 - NA Spring 2023-24 9 / 46

Eigenvalue Problem: Power Method (contd.)

Example:
A given matrix may have unique or more than one dominant (but
distinct or repeated) eigenvalues as illustrated here.
The matrix  
1 0 0
A= 0 −2 1
0 0 −1
has eigenvalues 1, −1, and −2.
Take λ1 = −2, λ2 = −1, and λ3 = 1.
The matrix A has a unique dominant eigenvalue, which is −2 as
this is the largest in absolute value, of all eigenvalues.
MA 214 - NA Spring 2023-24 9 / 46
Eigenvalue Problem: Power Method (contd.)

Example:
The matrix  
1 3 4

B= 0 2 1
0 0 −2

MA 214 - NA Spring 2023-24 10 / 46

Eigenvalue Problem: Power Method (contd.)

Example:
The matrix  
1 3 4

B= 0 2 1
0 0 −2
has eigenvalues 1, −2, 2.

MA 214 - NA Spring 2023-24 10 / 46

Eigenvalue Problem: Power Method (contd.)

Example:
The matrix  
1 3 4

B= 0 2 1
0 0 −2
has eigenvalues 1, −2, 2. According to our definition, the matrix B
has two dominant eigenvalues. They are −2 and 2.

MA 214 - NA Spring 2023-24 10 / 46

Eigenvalue Problem: Power Method (contd.)

Assume that an n × n matrix A has real eigenvalues λ1 , λ2 , · · · , and λn

with the following properties:

MA 214 - NA Spring 2023-24 11 / 46

Eigenvalue Problem: Power Method (contd.)

Assume that an n × n matrix A has real eigenvalues λ1 , λ2 , · · · , and λn

with the following properties:
1 The eigenvalues are such that
|λ1 | > |λ2 | ≥ |λ3 | ≥ · · · ≥ |λn |

MA 214 - NA Spring 2023-24 11 / 46

Eigenvalue Problem: Power Method (contd.)

Assume that an n × n matrix A has real eigenvalues λ1 , λ2 , · · · , and λn

with the following properties:
1 The eigenvalues are such that
|λ1 | > |λ2 | ≥ |λ3 | ≥ · · · ≥ |λn |

2 The corresponding eigenvectors v1 , v2 , · · · , vn are real and forms a

basis for Rn .

MA 214 - NA Spring 2023-24 11 / 46

Eigenvalue Problem: Power Method (contd.)

Assume that an n × n matrix A has real eigenvalues λ1 , λ2 , · · · , and λn

with the following properties:
1 The eigenvalues are such that
|λ1 | > |λ2 | ≥ |λ3 | ≥ · · · ≥ |λn |

2 The corresponding eigenvectors v1 , v2 , · · · , vn are real and forms a

basis for Rn .
Therefore, any vector v ∈ Rn can be written as
v = c1 v1 + c2 v2 + · · · + cn vn ,
for some c1 , c2 , · · · , cn ∈ R.
MA 214 - NA Spring 2023-24 11 / 46
Eigenvalue Problem: Power Method (contd.)
1 The eigenvalues are such that
|λ1 | > |λ2 | ≥ |λ3 | ≥ · · · ≥ |λn |.
2 The corresponding eigenvectors v1 , v2 , · · · , vn are real and forms a
basis for Rn .
Therefore, any vector v ∈ Rn can be written as
v = c1 v1 + c2 v2 + · · · + cn vn ,
for some c1 , c2 , · · · , cn ∈ R.
Pre-multiply by A and put Avi = λi vi , i = 1, · · · , n.

MA 214 - NA Spring 2023-24 12 / 46

Eigenvalue Problem: Power Method (contd.)
1 The eigenvalues are such that
|λ1 | > |λ2 | ≥ |λ3 | ≥ · · · ≥ |λn |.
2 The corresponding eigenvectors v1 , v2 , · · · , vn are real and forms a
basis for Rn .
Therefore, any vector v ∈ Rn can be written as
v = c1 v1 + c2 v2 + · · · + cn vn ,
for some c1 , c2 , · · · , cn ∈ R.
Pre-multiply by A and put Avi = λi vi , i = 1, · · · , n.
Av = c1 λ1 v1 + · · · + cn λn vn
MA 214 - NA Spring 2023-24 12 / 46
Eigenvalue Problem: Power Method (contd.)
Therefore, any vector v ∈ Rn can be written as
v = c1 v1 + c2 v2 + · · · + cn vn ,
for some c1 , c2 , · · · , cn ∈ R.
Pre-multiply by A and put Avi = λi vi , i = 1, · · · , n.
Av = c1 λ1 v1 + · · · +cn λ n vn   
λ2 λn
= λ1 c1 v1 + c2 v2 + · · · + cn vn .
λ1 λ1

MA 214 - NA Spring 2023-24 13 / 46

Eigenvalue Problem: Power Method (contd.)
Therefore, any vector v ∈ Rn can be written as
v = c1 v1 + c2 v2 + · · · + cn vn ,
for some c1 , c2 , · · · , cn ∈ R.
Pre-multiply by A and put Avi = λi vi , i = 1, · · · , n.
Av = c1 λ1 v1 + · · · +cn λ n vn   
λ2 λn
= λ1 c1 v1 + c2 v2 + · · · + cn vn .
λ1 λ1
Pre-multiply by A again and simplify to get

MA 214 - NA Spring 2023-24 13 / 46

Eigenvalue Problem: Power Method (contd.)
Therefore, any vector v ∈ Rn can be written as
v = c1 v1 + c2 v2 + · · · + cn vn ,
for some c1 , c2 , · · · , cn ∈ R.
Pre-multiply by A and put Avi = λi vi , i = 1, · · · , n.
Av = c1 λ1 v1 + · · · +cn λ n vn   
λ2 λn
= λ1 c1 v1 + c2 v2 + · · · + cn vn .
λ1 λ1
Pre-multiply by A again and simplify to get
 2  2 !
λ2 λn
A2 v = λ21 c1 v1 + c2 v2 + · · · + cn vn
λ1 λ1
MA 214 - NA Spring 2023-24 13 / 46
Eigenvalue Problem: Power Method (contd.)
Pre-multiply by A and put Avi = λi vi , i = 1, · · · , n.
Av = c1 λ1 v1 + · · · +cn λ
n vn   
λ2 λn
= λ1 c1 v1 + c2 v2 + · · · + cn vn .
λ1 λ1
Pre-multiply by A again and simplify to get
 2  2 !
λ 2 λn
A2 v = λ21 c1 v1 + c2 v 2 + · · · + cn vn
λ1 λ1
······
······

MA 214 - NA Spring 2023-24 14 / 46

Eigenvalue Problem: Power Method (contd.)
Pre-multiply by A and put Avi = λi vi , i = 1, · · · , n.
Av = c1 λ1 v1 + · · · +cn λ
n vn   
λ2 λn
= λ1 c1 v1 + c2 v2 + · · · + cn vn .
λ1 λ1
Pre-multiply by A again and simplify to get
 2  2 !
λ 2 λn
A2 v = λ21 c1 v1 + c2 v 2 + · · · + cn vn
λ1 λ1
······
······
 k  k !
λ2 λn
Ak v = λk1 c1 v1 + c2 v2 + · · · + cn vn
λ1 λ1
MA 214 - NA Spring 2023-24 14 / 46
Eigenvalue Problem: Power Method (contd.)

Using the assumption |λ1 | > |λ2 | ≥ |λ3 | ≥ · · · ≥ |λn |, we get

MA 214 - NA Spring 2023-24 15 / 46

Eigenvalue Problem: Power Method (contd.)

Using the assumption |λ1 | > |λ2 | ≥ |λ3 | ≥ · · · ≥ |λn |, we get

λj
< 1, j = 2, · · · , n.
λ1

MA 214 - NA Spring 2023-24 15 / 46

Eigenvalue Problem: Power Method (contd.)

Using the assumption |λ1 | > |λ2 | ≥ |λ3 | ≥ · · · ≥ |λn |, we get

λj
< 1, j = 2, · · · , n.
λ1
Using this in the expression !
 k  k
λ2 λn
Ak v = λk1 c1 v1 + c2 v2 + · · · + cn vn
λ1 λ1

MA 214 - NA Spring 2023-24 15 / 46

Eigenvalue Problem: Power Method (contd.)

Using the assumption |λ1 | > |λ2 | ≥ |λ3 | ≥ · · · ≥ |λn |, we get

λj
< 1, j = 2, · · · , n.
λ1
Using this in the expression !
 k  k
λ2 λn
Ak v = λk1 c1 v1 + c2 v2 + · · · + cn vn
λ1 λ1
Ak v
⇒ lim k = c1 v1 .
k→∞ λ
1
For c1 ̸= 0, the RHS of the above equation is a scalar multiple of the
eigenvector.
MA 214 - NA Spring 2023-24 15 / 46
Eigenvalue Problem: Power Method (contd.)
Using this in the expression
 k  k !
λ2 λn
Ak v = λk1 c1 v1 + c2 v2 + · · · + cn vn
λ1 λ1
Ak v
⇒ lim = c1 v1 .
k→∞ λk
1
For c1 ̸= 0, the RHS of the above equation is a scalar multiple of the
eigenvector.
Also, from the above expression for Ak v, we see that
(Ak+1 v)i
lim = λ1 ,
k→∞ (Ak v)i

where i is any non-zero component of the corresponding vectors.

MA 214 - NA Spring 2023-24 16 / 46
Eigenvalue Problem: Power Method (contd.)

The power method generates two sequences {µk } and {x(k) } using the
results
Ak v (Ak+1 v)i
lim = c1 v1 , lim = λ1
k→∞ λk k→∞ (Ak v)i
1

MA 214 - NA Spring 2023-24 17 / 46

Eigenvalue Problem: Power Method (contd.)

The power method generates two sequences {µk } and {x(k) } using the
results
Ak v (Ak+1 v)i
lim = c1 v1 , lim = λ1
k→∞ λk k→∞ (Ak v)i
1
that converges to the dominant eigenvalue λ1 and the corresponding
eigen vectors v1 , respectively.

MA 214 - NA Spring 2023-24 17 / 46

Eigenvalue Problem: Power Method (contd.)
We will now give the method of generating these two sequences.

MA 214 - NA Spring 2023-24 18 / 46

Eigenvalue Problem: Power Method (contd.)
We will now give the method of generating these two sequences.
Step 1: Choose a vector x(0) arbitrarily and set
y(1) = Ax(0) .

MA 214 - NA Spring 2023-24 18 / 46

Eigenvalue Problem: Power Method (contd.)
We will now give the method of generating these two sequences.
Step 1: Choose a vector x(0) arbitrarily and set
y(1) = Ax(0) .
Step:2 Let
n o
(1)
J = min j ∈ {1, 2, . . . , n} ∥y ∥∞ = |yj | .
(1)

Define
(1)
µ1 = yJ and

MA 214 - NA Spring 2023-24 18 / 46

Eigenvalue Problem: Power Method (contd.)
We will now give the method of generating these two sequences.
Step 1: Choose a vector x(0) arbitrarily and set
y(1) = Ax(0) .
Step:2 Let
n o
(1)
J = min j ∈ {1, 2, . . . , n} ∥y ∥∞ = |yj | .
(1)

Define
(1) y(1)
µ1 = yJ and x(1) = .
µ1

MA 214 - NA Spring 2023-24 18 / 46

Eigenvalue Problem: Power Method (contd.)
We will now give the method of generating these two sequences.
Step 1: Choose a vector x(0) arbitrarily and set
y(1) = Ax(0) .
Step:2 Let
n o
(1)
J = min j ∈ {1, 2, . . . , n} ∥y ∥∞ = |yj | .
(1)

Define
(1) y(1)
µ1 = yJ and x(1) = .
µ1
From x(1) , we can obtain µ2 and x(2) and so on.
MA 214 - NA Spring 2023-24 18 / 46
Eigenvalue Problem: Power Method (contd.)

Iterative Procedure for Power Method

Choose the initial vector x(0) arbitrarily;

MA 214 - NA Spring 2023-24 19 / 46

Eigenvalue Problem: Power Method (contd.)

Iterative Procedure for Power Method

Choose the initial vector x(0) arbitrarily;

For k = 0, 1, · · · , define
y(k+1) = Ax(k) .
Let n o
(k+1)
J = min j ∈ {1, 2, . . . , n} ∥y (k+1)
∥∞ = |yj | .
Define the two sequences {µk } and {x(k) } using the formulas
(k+1) y(k+1)
µk+1 = yJ , x(k+1) = .
µk+1
MA 214 - NA Spring 2023-24 19 / 46
Eigenvalue Problem: Power Method (contd.)

Example:
Consider the matrix
 
3 0 0
A =  −4 6 2 
16 −15 −5

MA 214 - NA Spring 2023-24 20 / 46

Eigenvalue Problem: Power Method (contd.)

Example:
Consider the matrix
 
3 0 0
A =  −4 6 2 
16 −15 −5
The eigenvalues of this matrix are
λ1 = 3, λ2 = 1, and λ3 = 0

MA 214 - NA Spring 2023-24 20 / 46

Eigenvalue Problem: Power Method (contd.)

Example:
Consider the matrix
 
3 0 0
A =  −4 6 2 
16 −15 −5
The eigenvalues of this matrix are
λ1 = 3, λ2 = 1, and λ3 = 0
Let us take the corresponding eigenvectors as
v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and v3 = (0, 1, −3)T .

MA 214 - NA Spring 2023-24 20 / 46

Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are
λ1 = 3, λ2 = 1, and λ3 = 0
The corresponding eigen vectors are
v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and v3 = (0, 1, −3)T .
Initial Guess 1: Let us take x(0) = (1, 0.5, 0.25)T .

MA 214 - NA Spring 2023-24 21 / 46

Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are
λ1 = 3, λ2 = 1, and λ3 = 0
The corresponding eigen vectors are
v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and v3 = (0, 1, −3)T .
Initial Guess 1: Let us take x(0) = (1, 0.5, 0.25)T .
Iteration No: 1
y(1) = Ax(0) = (3.000000, −0.500000, 7.250000, )T
µ1 = 7.250000
y(1)
x(1)
= = (0.413793, −0.068966, 1.000000)T
µ1
MA 214 - NA Spring 2023-24 21 / 46
Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 2
y(2) = Ax(1) = (1.241379, −0.068966, 2.655172)T
µ2 = 2.655172
y(2)
x(2) = = (0.467532, −0.025974, 1.000000)T
µ2

MA 214 - NA Spring 2023-24 22 / 46

Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 2
y(2) = Ax(1) = (1.241379, −0.068966, 2.655172)T
µ2 = 2.655172
y(2)
x(2) = = (0.467532, −0.025974, 1.000000)T
µ2
Iteration No: 3
y(3) = Ax(2) = (1.402597, −0.025974, 2.870130)T
µ3 = 2.870130
y(3)
x(3)
= = (0.488688, −0.009050, 1.000000)T
µ3
MA 214 - NA Spring 2023-24 22 / 46
Eigenvalue Problem: Power Method (contd.)

The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The

corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 4 y(4) = Ax(3) = (1.466063, −0.009050, 2.954751)T
µ4 = 2.954751
y(4)
x(4) = = (0.496172, −0.003063, 1.000000)T
µ4

MA 214 - NA Spring 2023-24 23 / 46

Eigenvalue Problem: Power Method (contd.)

The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The

corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 4 y(4) = Ax(3) = (1.466063, −0.009050, 2.954751)T
µ4 = 2.954751
y(4)
x(4) = = (0.496172, −0.003063, 1.000000)T
µ4
Iteration No: 5
y(5) = Ax(4) = (1.488515, −0.003063, 2.984686)T
µ5 = 2.984686
y(5)
x(5) = = (0.498717, −0.001026, 1.000000)T
µ5
MA 214 - NA Spring 2023-24 23 / 46
Eigenvalue Problem: Power Method (contd.)

The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The

corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 6
y(6) = Ax(5) = (1.496152, −0.001026, 2.994869)T
µ6 = 2.994869
y(6)
x(6) = = (0.499572, −0.000343, 1.000000)T
µ6

MA 214 - NA Spring 2023-24 24 / 46

Eigenvalue Problem: Power Method (contd.)

The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The

corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 6
y(6) = Ax(5) = (1.496152, −0.001026, 2.994869)T
µ6 = 2.994869
y(6)
x(6) = = (0.499572, −0.000343, 1.000000)T
µ6
Iteration No: 7
y(7) = Ax(6) = (1.498715, −0.000343, 2.998287)T
µ7 = 2.998287
y(7)
x(7) = = (0.499857, −0.000114, 1.000000)T
µ7
MA 214 - NA Spring 2023-24 24 / 46
Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 8
y(8) = Ax(7) = (1.499571, −0.000114, 2.999429)T
µ8 = 2.999429
y(8)
x(8) = = (0.499952, −0.000038, 1.000000)T
µ8

MA 214 - NA Spring 2023-24 25 / 46

Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 8
y(8) = Ax(7) = (1.499571, −0.000114, 2.999429)T
µ8 = 2.999429
y(8)
x(8) = = (0.499952, −0.000038, 1.000000)T
µ8
Iteration No: 9
y(9) = Ax(8) = (1.499857, −0.000038, 2.999809)T
µ9 = 2.999809
y(9)
x(9) = = (0.499984, −0.000013, 1.000000)T
µ9
MA 214 - NA Spring 2023-24 25 / 46
Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 10
y(10) = Ax(9) = (1.499952, −0.000013, 2.999936)T
µ10 = 2.999936
y(10)
x(10) = = (0.499995, −0.000004, 1.000000)T
µ10

MA 214 - NA Spring 2023-24 26 / 46

Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 10
y(10) = Ax(9) = (1.499952, −0.000013, 2.999936)T
µ10 = 2.999936
y(10)
x(10) = = (0.499995, −0.000004, 1.000000)T
µ10
We observe that the sequence {µk } converges to the eigenvalue λ1 = 3
1
and the sequence {x(k) } converges to (0.5, 0, 1) = v1 .
2
MA 214 - NA Spring 2023-24 26 / 46
Eigenvalue Problem: Power Method (contd.)

Example: Consider the matrix

 
1 3 4
B=0 2 1 ,
0 0 −2

which has eigenvalues 1, −2, and 2. Clearly, the matrix B has two
dominant eigenvalues, namely, −2 and 2.

MA 214 - NA Spring 2023-24 27 / 46

Eigenvalue Problem: Power Method (contd.)

We start with an initial guess x(0) = (1, 1, 1) and the five iterations
generated using power method are given below:

MA 214 - NA Spring 2023-24 28 / 46

Eigenvalue Problem: Power Method (contd.)

We start with an initial guess x(0) = (1, 1, 1) and the five iterations
generated using power method are given below:
Iteration No: 1
y(1) = Ax(0) = (8.000000, 3.000000, −2.000000, )T
µ1 = 8.000000
y(1)
x(1) = = (1.000000, 0.375000, −0.250000, )T
µ1

MA 214 - NA Spring 2023-24 28 / 46

Eigenvalue Problem: Power Method (contd.)

We start with an initial guess x(0) = (1, 1, 1) and the five iterations
generated using power method are given below:
Iteration No: 1
y(1) = Ax(0) = (8.000000, 3.000000, −2.000000, )T
µ1 = 8.000000
y(1)
x(1) = = (1.000000, 0.375000, −0.250000, )T
µ1
Iteration No: 2 (2)
y = Ax(1) = (1.125000, 0.500000, 0.500000, )T
µ2 = 1.125000
y(2)
x(2) = = (1.000000, 0.444444, 0.444444, )T
µ2

MA 214 - NA Spring 2023-24 28 / 46

Eigenvalue Problem: Power Method (contd.)

Iteration No: 3

y(3) = Ax(2) = (4.111111, 1.333333, −0.888889, )T

µ3 = 4.111111
y(3)
x(3)
= = (1.000000, 0.324324, −0.216216, )T
µ3

MA 214 - NA Spring 2023-24 29 / 46

Eigenvalue Problem: Power Method (contd.)

Iteration No: 3

y(3) = Ax(2) = (4.111111, 1.333333, −0.888889, )T

µ3 = 4.111111
y(3)
x(3)
= = (1.000000, 0.324324, −0.216216, )T
µ3
Iteration No: 4

y(4) = Ax(3) = (1.108108, 0.432432, 0.432432, )T

µ4 = 1.108108
y(4)
x(4) = = (1.000000, 0.390244, 0.390244, )T
µ4

MA 214 - NA Spring 2023-24 29 / 46

Eigenvalue Problem: Power Method (contd.)

It is observed that the sequence oscillates even after 1000 iterations as

shown below:
Iteration No: 997
y(997) = Ax(996) = (3.625000, 1.125000, −0.750000, )T
µ997 = 3.625000
y(997)
x(997) = = (1.000000, 0.310345, −0.206897, )T
µ997

MA 214 - NA Spring 2023-24 30 / 46

Eigenvalue Problem: Power Method (contd.)

It is observed that the sequence oscillates even after 1000 iterations as

shown below:
Iteration No: 997
y(997) = Ax(996) = (3.625000, 1.125000, −0.750000, )T
µ997 = 3.625000
y(997)
x(997) = = (1.000000, 0.310345, −0.206897, )T
µ997
Iteration No: 998
y(998) = Ax(997) = (1.103448, 0.413793, 0.413793, )T
µ998 = 1.103448
y(998)
x(998) = = (1.000000, 0.375000, 0.375000, )T
µ998
MA 214 - NA Spring 2023-24 30 / 46
Eigenvalue Problem: Power Method (contd.)

Iteration No: 999

y(999) = Ax(998) = (3.625000, 1.125000, −0.750000, )T

µ999 = 3.625000
y(999)
x(999) = = (1.000000, 0.310345, −0.206897, )T
µ999

MA 214 - NA Spring 2023-24 31 / 46

Eigenvalue Problem: Power Method (contd.)

Iteration No: 999

y(999) = Ax(998) = (3.625000, 1.125000, −0.750000, )T

µ999 = 3.625000
y(999)
x(999) = = (1.000000, 0.310345, −0.206897, )T
µ999
Iteration No: 1000
y(1000) = Ax(999) = (1.103448, 0.413793, 0.413793, )T
µ1000 = 1.103448
y(1000)
x(1000) = = (1.000000, 0.375000, 0.375000, )T
µ1000

and so on.
MA 214 - NA Spring 2023-24 31 / 46
Eigenvalue Problem: Power Method (contd.)
Theorem (Convergence Theorem for Power method)
Hypothesis: Let A be an n × n matrix with real eigenvalues having the following
properties:

MA 214 - NA Spring 2023-24 32 / 46

Eigenvalue Problem: Power Method (contd.)
Theorem (Convergence Theorem for Power method)
Hypothesis: Let A be an n × n matrix with real eigenvalues having the following
properties:
(H1) A has a unique dominant eigenvalue λ1 which is a simple eigenvalue.
That is,
|λ1 | > |λ2 | ≥ · · · ≥ |λn |,
where λ1 , λ2 , · · · , λn are the eigenvalues of A (repeated according to their
algebraic multiplicities).

MA 214 - NA Spring 2023-24 32 / 46

Eigenvalue Problem: Power Method (contd.)
Theorem (Convergence Theorem for Power method)
Hypothesis: Let A be an n × n matrix with real eigenvalues having the following
properties:
(H1) A has a unique dominant eigenvalue λ1 which is a simple eigenvalue.
That is,
|λ1 | > |λ2 | ≥ · · · ≥ |λn |,
where λ1 , λ2 , · · · , λn are the eigenvalues of A (repeated according to their
algebraic multiplicities).
(H2) A has n linearly independent real eigenvectors, vi , i = 1, · · · , n.

MA 214 - NA Spring 2023-24 32 / 46

Eigenvalue Problem: Power Method (contd.)
Theorem (Convergence Theorem for Power method)
Hypothesis: Let A be an n × n matrix with real eigenvalues having the following
properties:
(H1) A has a unique dominant eigenvalue λ1 which is a simple eigenvalue.
That is,
|λ1 | > |λ2 | ≥ · · · ≥ |λn |,
where λ1 , λ2 , · · · , λn are the eigenvalues of A (repeated according to their
algebraic multiplicities).
(H2) A has n linearly independent real eigenvectors, vi , i = 1, · · · , n.
(H3) An initial guess x(0) ∈ Rn be chosen such that
Xn
x(0) = cj vj ,
j=1
[
∞
for some scalars c1 , c2 , · · · , cn ∈ R with c1 ̸= 0 and x(0) ∈
/ KerAk .
k=1
MA 214 - NA Spring 2023-24 32 / 46
Eigenvalue Problem: Power Method (contd.)
Theorem (Convergence Theorem for Power method)
Conclusion: Then, in the power method
1 the sequence {µk } converges to the dominant eigenvalue λ1 and
2 a subsequence of {x(k) } converges to an eigenvector of λ1 .
Proof is left as a reading exercise.
 v1

 either +

 ∥v1 ∥∞




 v1
lim x(k+1) = or − =⇒ {x(k) } converges up to a subsequence
 ∥v1 ∥∞
k→∞ 






 or oscillates between
the above two vectors
MA 214 - NA Spring 2023-24 33 / 46
Eigenvalue Problem: Power Method (contd.)

Hypothesis 3:
An initial guess x(0) ∈ Rn be chosen such that
X n
(0)
x = cj vj ,
j=1
∞
[
for some scalars c1 , c2 , · · · , cn ∈ R with c1 ̸= 0 and x
(0)
∈
/ KerAk .
k=1

MA 214 - NA Spring 2023-24 34 / 46

Eigenvalue Problem: Power Method (contd.)

Hypothesis 3:
An initial guess x(0) ∈ Rn be chosen such that
X n
(0)
x = cj vj ,
j=1
∞
[
for some scalars c1 , c2 , · · · , cn ∈ R with c1 ̸= 0 and x
(0)
∈
/ KerAk .
k=1
Consider the matrix
 
3 0 0
A =  −4 6 2 ,
16 −15 −5
MA 214 - NA Spring 2023-24 34 / 46
Eigenvalue Problem: Power Method (contd.)
Consider the matrix  
3 0 0
A =  −4 6 2 ,
16 −15 −5
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are
v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and v3 = (0, 1, −3)T .
Let us take x(0) = (0, 0.5, 0.25)T .

MA 214 - NA Spring 2023-24 35 / 46

Eigenvalue Problem: Power Method (contd.)
Consider the matrix  
3 0 0
A =  −4 6 2 ,
16 −15 −5
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are
v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and v3 = (0, 1, −3)T .
Let us take x(0) = (0, 0.5, 0.25)T .
Iteration No: 1
y(1) = Ax(0) = (0.000000, 3.500000, −8.750000)T
µ1 = −8.750000
y(1)
x(1) = = (0.000000, −0.400000, 1.000000)T
µ1
MA 214 - NA Spring 2023-24 35 / 46
Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .

MA 214 - NA Spring 2023-24 36 / 46

Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 2
y(2) = Ax(1) = (0.000000, −0.400000, 1.000000)T
µ2 = 1.000000
y(2)
x(2) = = (0.000000, −0.400000, 1.000000)T
µ2

MA 214 - NA Spring 2023-24 36 / 46

Eigenvalue Problem: Power Method (contd.)
The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The
corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 2
y(2) = Ax(1) = (0.000000, −0.400000, 1.000000)T
µ2 = 1.000000
y(2)
x(2) = = (0.000000, −0.400000, 1.000000)T
µ2
Iteration No: 3
y(3) = Ax(2) = (0.000000, −0.400000, 1.000000)T
µ3 = 1.000000
y(3)
x(3) = = (0.000000, −0.400000, 1.000000)T
µ3
MA 214 - NA Spring 2023-24 36 / 46
Eigenvalue Problem: Power Method (contd.)

The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The

corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .

MA 214 - NA Spring 2023-24 37 / 46

Eigenvalue Problem: Power Method (contd.)

The eigenvalues of this matrix are λ1 = 3, λ2 = 1 and λ3 = 0. The

corresponding eigen vectors are v1 = (1, 0, 2)T , v2 = (0, 2, −5)T and
v3 = (0, 1, −3)T .
Iteration No: 4

y(4) = Ax(3) = (0.000000, −0.400000, 1.000000)T

µ4 = 1.000000
y(4)
x(4) = = (0.000000, −0.400000, 1.000000)T
µ4

MA 214 - NA Spring 2023-24 37 / 46

Eigenvalue Problem: Power Method (contd.)
Note that in the initial guess x(0) = (0, 0.5, 0.25)T of the above example,
the first coordinate is zero.

MA 214 - NA Spring 2023-24 38 / 46

Eigenvalue Problem: Power Method (contd.)
Note that in the initial guess x(0) = (0, 0.5, 0.25)T of the above example,
the first coordinate is zero.
Therefore c1 in the representation
Xn
(0)
x = cj vj ,
j=1
has to be zero.

MA 214 - NA Spring 2023-24 38 / 46

Eigenvalue Problem: Power Method (contd.)
Note that in the initial guess x(0) = (0, 0.5, 0.25)T of the above example,
the first coordinate is zero.
Therefore c1 in the representation
Xn
(0)
x = cj vj ,
j=1
has to be zero.
This is because the corresponding eigen vectors are v1 = (1, 0, 2)T ,
v2 = (0, 2, −5)T and v3 = (0, 1, −3)T , and therefore

MA 214 - NA Spring 2023-24 38 / 46

Eigenvalue Problem: Power Method (contd.)
Note that in the initial guess x(0) = (0, 0.5, 0.25)T of the above example,
the first coordinate is zero.
Therefore c1 in the representation
Xn
(0)
x = cj vj ,
j=1
has to be zero.
This is because the corresponding eigen vectors are v1 = (1, 0, 2)T ,
v2 = (0, 2, −5)T and v3 = (0, 1, −3)T , and therefore
(0, 0.5, 0.25)T = c1 (1, 0, 2)T + c2 (0, 2, −5)T + c3 (0, 1, −3)T
⇒
c1 = 0
MA 214 - NA Spring 2023-24 38 / 46
Eigenvalue Problem: Power Method (contd.)

Thus, the relation

 k  k !
λ2 λn
Ak v = λk1 c1 v1 + c2 v2 + · · · + cn vn
λ1 λ1

MA 214 - NA Spring 2023-24 39 / 46

Eigenvalue Problem: Power Method (contd.)

Thus, the relation

 k  k !
λ2 λn
Ak v = λk1 c1 v1 + c2 v2 + · · · + cn vn
λ1 λ1

reduces to
 k  k !
λ3 λn
Ak v = λk2 c2 v2 + c3 v3 + · · · + cn vn .
λ2 λ2

MA 214 - NA Spring 2023-24 39 / 46

Eigenvalue Problem: Power Method (contd.)

Thus, the relation

 k  k !
λ2 λn
Ak v = λk1 c1 v1 + c2 v2 + · · · + cn vn
λ1 λ1

reduces to
 k  k !
λ3 λn
Ak v = λk2 c2 v2 + c3 v3 + · · · + cn vn .
λ2 λ2

This makes the iteration to converge to λ2 , which is the next dominant

eigenvalue.
MA 214 - NA Spring 2023-24 39 / 46
Eigenvalue Problem: Power Method (contd.)
Disadvantages of power method

1 The Power method requires at the beginning that the matrix has
only one dominant eigenvalue, and this information is generally
unavailable.
2 Even when there is only one dominant eigenvalue, it is not clear
how to choose the initial guess x(0) such that it has a non-zero
component (c1 in the notation of the theorem) along the
eigenvector v1 .
Note that in the above example, all the hypothesis of power method are
satisfied. Now let us ask the question
“What happens when any of these hypotheses is violated?”
MA 214 - NA Spring 2023-24 40 / 46
Eigenvalue Problem: Inverse Power Method

Recall a result from Linear Algebra:

MA 214 - NA Spring 2023-24 41 / 46

Eigenvalue Problem: Inverse Power Method

Recall a result from Linear Algebra:

Theorem
If λ is an eigenvalue of an invertible matrix A, then λ−1 is an eigenvalue
of A−1 .

MA 214 - NA Spring 2023-24 41 / 46

Eigenvalue Problem: Inverse Power Method

Recall a result from Linear Algebra:

Theorem
If λ is an eigenvalue of an invertible matrix A, then λ−1 is an eigenvalue
of A−1 .
Assume: The eigenvalues of an invertible n × n matrix A (after a
re-arrangement) are such that
|λ1 | ≥ |λ2 | ≥ . . . ≥ |λn−1 | > |λn |.

MA 214 - NA Spring 2023-24 41 / 46

Eigenvalue Problem: Inverse Power Method

Recall a result from Linear Algebra:

Theorem
If λ is an eigenvalue of an invertible matrix A, then λ−1 is an eigenvalue
of A−1 .
Assume: The eigenvalues of an invertible n × n matrix A (after a
re-arrangement) are such that
|λ1 | ≥ |λ2 | ≥ . . . ≥ |λn−1 | > |λn |.
Then for A−1 , we have
1 1 1 1
> ≥ ... ≥ ≥ .
|λn | |λn−1 | |λ2 | |λ1 |
MA 214 - NA Spring 2023-24 41 / 46
Eigenvalue Problem: Inverse Power Method

Assume: The eigenvalues of an invertible n × n matrix A (after a

re-arrangement) are such that
|λ1 | ≥ |λ2 | ≥ . . . ≥ |λn−1 | > |λn |.
Then for A−1 , we have
1 1 1 1
> ≥ ... ≥ ≥ .
|λn | |λn−1 | |λ2 | |λ1 |
1
Power method may be used to approximate λn and the corresponding
eigenvector vn .

MA 214 - NA Spring 2023-24 42 / 46

Eigenvalue Problem: Inverse Power Method

Assume: The eigenvalues of an invertible n × n matrix A (after a

re-arrangement) are such that
|λ1 | ≥ |λ2 | ≥ . . . ≥ |λn−1 | > |λn |.
Then for A−1 , we have
1 1 1 1
> ≥ ... ≥ ≥ .
|λn | |λn−1 | |λ2 | |λ1 |
1
Power method may be used to approximate λn and the corresponding
eigenvector vn .
Hence, we can approximate the smallest (in absolute value) eigenvalue
of A.
MA 214 - NA Spring 2023-24 42 / 46
Eigenvalue Problem: Inverse Power Method
Iterative Procedure for Inverse Power Method
We choose the initial vector x(0) arbitrarily.

MA 214 - NA Spring 2023-24 43 / 46

Eigenvalue Problem: Inverse Power Method
Iterative Procedure for Inverse Power Method
We choose the initial vector x(0) arbitrarily.
For k = 0, 1, · · · , let
y(k+1) = A−1 x(k)

MA 214 - NA Spring 2023-24 43 / 46

Eigenvalue Problem: Inverse Power Method
Iterative Procedure for Inverse Power Method
We choose the initial vector x(0) arbitrarily.
For k = 0, 1, · · · , let
y(k+1) = A−1 x(k) =⇒ Ay(k+1) = x(k)

MA 214 - NA Spring 2023-24 43 / 46

Eigenvalue Problem: Inverse Power Method
Iterative Procedure for Inverse Power Method
We choose the initial vector x(0) arbitrarily.
For k = 0, 1, · · · , let
y(k+1) = A−1 x(k) =⇒ Ay(k+1) = x(k) (LU factorization).

MA 214 - NA Spring 2023-24 43 / 46

Eigenvalue Problem: Inverse Power Method
Iterative Procedure for Inverse Power Method
We choose the initial vector x(0) arbitrarily.
For k = 0, 1, · · · , let
y(k+1) = A−1 x(k) =⇒ Ay(k+1) = x(k) (LU factorization).
Define
 (k+1)
J = min j | ∥y(k+1) ∥∞ = |yj |, j = 1, 2, . . . , n .

MA 214 - NA Spring 2023-24 43 / 46

Eigenvalue Problem: Inverse Power Method
Iterative Procedure for Inverse Power Method
We choose the initial vector x(0) arbitrarily.
For k = 0, 1, · · · , let
y(k+1) = A−1 x(k) =⇒ Ay(k+1) = x(k) (LU factorization).
Define
 (k+1)
J = min j | ∥y(k+1) ∥∞ = |yj |, j = 1, 2, . . . , n .
Finally, define two sequences {µk } and {x(k) } using the formulas
(k+1) y(k+1)
µk+1 = yJ , x(k+1) = .
µk+1
MA 214 - NA Spring 2023-24 43 / 46
Eigenvalue Problem: Inverse Power Method
If these sequences converge (with x(0) is such that c1 ̸= 0),

where do they converge?

MA 214 - NA Spring 2023-24 44 / 46

Eigenvalue Problem: Inverse Power Method
If these sequences converge (with x(0) is such that c1 ̸= 0),

where do they converge?

µk → 1/λn

MA 214 - NA Spring 2023-24 44 / 46

Eigenvalue Problem: Inverse Power Method
If these sequences converge (with x(0) is such that c1 ̸= 0),

where do they converge?

µk → 1/λn
(possibly a subsequence) x(k) converges to an eigenvector of λn .

MA 214 - NA Spring 2023-24 44 / 46

Eigenvalue Problem: Inverse Power Method
Example: Consider
 
−2.7083 −2.6824 0.4543
A =  0.1913 0.7629 0.1007  .
−0.3235 −0.4052 5.0453

MA 214 - NA Spring 2023-24 45 / 46

Eigenvalue Problem: Inverse Power Method
Example: Consider
 
−2.7083 −2.6824 0.4543
A =  0.1913 0.7629 0.1007  .
−0.3235 −0.4052 5.0453

We have
   
1 0 0 −2.7083 − 2.6824 0.4543
L=  −0.0706 1 0  U=  0 0.5734 0.1328  .
0.1194 − 0.1479 1 0 0 5.0107

MA 214 - NA Spring 2023-24 45 / 46

Eigenvalue Problem: Inverse Power Method
Iterations: (approximately)
y(1) = (−2.14, 1.82, 0.21)T , µ1 = −2.14, x(1) = (1, −0.85, −0.10)T .

MA 214 - NA Spring 2023-24 46 / 46

Eigenvalue Problem: Inverse Power Method
Iterations: (approximately)
y(1) = (−2.14, 1.82, 0.21)T , µ1 = −2.14, x(1) = (1, −0.85, −0.10)T .
y(2) = (0.95, −1.35 − 0.07)T , µ2 = −1.35, x(2) = (−0.71, 1.00, 0.05)T .

MA 214 - NA Spring 2023-24 46 / 46

Eigenvalue Problem: Inverse Power Method
Iterations: (approximately)
y(1) = (−2.14, 1.82, 0.21)T , µ1 = −2.14, x(1) = (1, −0.85, −0.10)T .
y(2) = (0.95, −1.35 − 0.07)T , µ2 = −1.35, x(2) = (−0.71, 1.00, 0.05)T .
y(3) = (−1.36, 1.64, 0.05)T , µ3 = 1.64, x(3) = (−0.83, 1.00, 0.03)T .

MA 214 - NA Spring 2023-24 46 / 46

Eigenvalue Problem: Inverse Power Method
Iterations: (approximately)
y(1) = (−2.14, 1.82, 0.21)T , µ1 = −2.14, x(1) = (1, −0.85, −0.10)T .
y(2) = (0.95, −1.35 − 0.07)T , µ2 = −1.35, x(2) = (−0.71, 1.00, 0.05)T .
y(3) = (−1.36, 1.64, 0.05)T , µ3 = 1.64, x(3) = (−0.83, 1.00, 0.03)T .
y(4) = (−1.30, 1.63, 0.05)T , µ4 = 1.63, x(4) = (−0.80, 1.00, 0.03)T .
y(5) = (−1.31, 1.63, 0.05)T , µ5 = 1.63, x(5) = (−0.80, 1.00, 0.03)T .

MA 214 - NA Spring 2023-24 46 / 46

Eigenvalue Problem: Inverse Power Method
Iterations: (approximately)
y(1) = (−2.14, 1.82, 0.21)T , µ1 = −2.14, x(1) = (1, −0.85, −0.10)T .
y(2) = (0.95, −1.35 − 0.07)T , µ2 = −1.35, x(2) = (−0.71, 1.00, 0.05)T .
y(3) = (−1.36, 1.64, 0.05)T , µ3 = 1.64, x(3) = (−0.83, 1.00, 0.03)T .
y(4) = (−1.30, 1.63, 0.05)T , µ4 = 1.63, x(4) = (−0.80, 1.00, 0.03)T .
y(5) = (−1.31, 1.63, 0.05)T , µ5 = 1.63, x(5) = (−0.80, 1.00, 0.03)T .
Let us stop here and take
λ = 1/µ5 = 0.61
The smallest eigenvalue of A is approximately 0.6125.

A corresponding vector is approximately (−0.49, 0.61, 0.02)T , whereas we got

x(5) = (−0.80, 1.00, 0.03)T .
MA 214 - NA Spring 2023-24 46 / 46