Lec 21

Diagonalization

Diagonalization

Definition: A 𝑛 × 𝑛 matrix 𝐴 is said to be diagonalizable if there is a diagonal matrix 𝐷 such

that 𝐴 is similar to 𝐷. That is, if there is an invertible matrix 𝑃 such that 𝑃 𝐴𝑃 = 𝐷.

Example:

1 3 1 3 4 0
𝐴= is diagonalizable if 𝑃 = and 𝐷 = .
2 2 1 −2 0 −1

Theorem 4.23:
Let 𝐴 be a 𝑛 × 𝑛 matrix. A is diagonalizable iff 𝐴 has 𝑛 linearly independent eigen vectors.

Example 1:

3 −2 0
Let 𝐴 = −2 3 0.
0 0 5

Eigen values are 𝜆 = 1, 5, 5.

𝐸 (1) = 𝑠𝑝𝑎𝑛{(1,1,0)}, 𝐸 (5) = 𝑠𝑝𝑎𝑛{(−1,1,0), (0, 0, 1)}.

1 −1 0 1 0 0
𝑃= 1 1 0 D= 0 5 0
0 0 1 0 0 5

Verify 𝐴𝑃 = 𝑃𝐷.

Example 2:

0 1 0
Let 𝐴 = 0 0 1.
2 −5 4

Eigen values are 𝜆 = 1, 1, 2

𝐸 (1) = 𝑠𝑝𝑎𝑛{(1, 1, 1)} and 𝐸 (2) = 𝑠𝑝𝑎𝑛{(1, 2, 4)}.
Algebraic multiplicity of 𝜆 = 1 is 2 and GM is 1.
AM of 𝜆 = 2 is 1 =GM of 𝐸 .

There are only two L.I Eigen vectors and hence we can not find P and therefore, A is not diagonalizable.
Example 3:

𝑎 1 0 0 0
⎡0 𝑎 0 0 ⎤⎥
⎢ 1
Let 𝐴 = ⎢ 0 0 𝑎 1 0⎥
⎢0 0 0 𝑎 1⎥
⎣0 0 0 0 0⎦

Is A diagonalizable?

What are the eigen values of A and eigen space of A?

In short:
𝐴 is diagonalizable implies:

1. 𝑃 𝐴𝑃 = 𝐷, P - invertible matrix, D- Diagonal matrix.

2. The columns of 𝑃 are the linearly independent eigenvectors of 𝐴.

3. The diagonal entries of 𝐷 are the eigenvalues of 𝐴 corresponding to the columns (eigenvectors of
A) of P in the same order.

Question: In example 1, we obtained 𝑃 = [𝑝 , 𝑝 , 𝑝 ].

Is it necessary to check 𝑝 , 𝑝 , 𝑝 are linearly independent?

Answer:
Theorem 4.24:
Let 𝜆 , 𝜆 , … 𝜆 be distinct eigenvalues of a matrix 𝐴. If 𝐵 is a basis for the eigenspace 𝐸(𝜆 ), then
𝐵 = 𝐵 ∪ 𝐵 ∪ ⋯ ∪ 𝐵 is a linearly independent set.

Theorem 4.25: If 𝐴 is an 𝑛 × 𝑛 matrix with 𝑛 distinct eigen values then 𝐴 is diagonalizable.

Converse need not be true.

Example 2

Theorem 4.27: The Diagonalization Theorem

Let 𝐴 be an 𝑛 × 𝑛 matrix whose distinct eigenvalues are 𝜆 , λ , … , 𝜆 . Then the following are equivalent:

a. 𝐴 is diagonalizable.
b. The union 𝐵 of the bases of eigenspaces of 𝐴 contains 𝑛 vectors.
c. The algebraic multiplicity of each eigenvalue equals its geometric multiplicity.

Examples:

Example 2-----since for λ = 1, AM ≠GM, by Thm 4.27, it is not diagonalizable

Example 1 -----Diagonalizable (why?)

Remark:

If 𝐴 is diagonalizable, then there exists invertible matrix 𝑃 such that

𝐷 = 𝑃 𝐴𝑃.
⟹ 𝐴 = 𝑃𝐷𝑃 ⟹ 𝐴 = 𝑃 𝐷 𝑃 for any 𝑚 > 0.

Higher powers of 𝐴 can also be computed using diagonalization.