Leergedeelte 3.

2
Study Section 3.2
Diagonalisering van simmetriese matrikse
Diagonalization of symmetric matrices

1
 Uitkomste / Outcomes
O Begrippe: Simmetriese matriks, ortogonaal-
diagonaliseerbare matriks, spektraalontbinding van ‘n
matriks en projeksiematriks.
O Concepts: Symmetric matrix, orthogonal diagonalizable
matrix, spectral decomposition of a matrix and
projection matrix.
O Stelling 1: Enige twee eievektore van verskillende
eieruimtes van ‘n simmetriese matriks A is ortogonaal
(bl. {413}(427) [395]) (Bewys ingesluit).
O Theorem 1: Any two eigenvectors of different
eigenspaces of a symmetric matrix A are orthogonal (p.
{413} (427) [395]) (proof included).

2
 Uitkomste (vervolg) / Outcomes (continued)
O Stelling 2: ‘n Vierkantige matriks is ortogonaal
diagonaliseerbaar as en slegs as dit simmetries is (bl.
{414} (428)[396]).
O Theorem 2: S square matrix is orthogonally
diagonalizable if and only if it is symmetric (p.
{414}(428)[396]).
O Die Spektraalstelling vir simmetriese matrikse (Stelling
3, bl. {415}(429) [397]).
O Spectral Theorem for Symmetric Matrices (Theorem 3,
p. {415}(429)[397]).
O Berekeninge / Calculations.
O Eie bewyse. / Own proofs.
3
 Definisie / Definition
'n Matriks 𝐴 word simmetries genoem indien 𝐴𝑇 = 𝐴.
A matrix 𝐴 is called symmetric if 𝐴𝑇 = 𝐴.
Let op: / Note:
O As 𝐴 simmetries is, is 𝐴 noodwendig vierkantig.
O If 𝐴 is symmetric, 𝐴 is necessarily square.
O Die inskrywings op die hoofdiagonaal van 'n
simmetriese matriks kan enige waardes hê, maar die
ander inskrywings kom in pare weerskante van die
hoofdiagonaal voor.
O The entries on the main diagonal of a symmetric matrix
can take on any values, but the other entries appear in
pairs on opposite sides of the main diagonal.
4
 Stelling 1 / Theorem 1

O As 𝐴 simmetries is, dan is enige twee eievektore van

5
 Definisie / Definition
'n 𝑛 × 𝑛 Matriks 𝐴 word ortogonaal-diagonaliseerbaar
genoem as daar 'n ortogonale matriks 𝑃 (met 𝑃−1 =
𝑃𝑇 ) en 'n diagonaalmatriks 𝐷 bestaan só dat
An 𝑛 × 𝑛 matrix 𝐴 is called orthogonally diagonalizable
if there exists an orthogonal matrix 𝑃 (with 𝑃−1 = 𝑃𝑇 )
and a diagonal matrix 𝐷 such that

𝐴 = 𝑃𝐷𝑃𝑇 = 𝑃𝐷𝑃−1

6
 Stelling 2 / Theorem 2
'n 𝑛 × 𝑛 Matriks 𝐴 is ortogonaal-diagonaliseerbaar as
en slegs as 𝐴 'n simmetriese matriks is.
An 𝑛 × 𝑛 matrix 𝐴 is orthogonally diagonalizable if and
only if 𝐴 is a symmetric matrix.
Bewys van / Proof of (⟹):
Laat 𝐴 ortogonaal-diagonaliseerbaar wees, dws 𝑃𝐷𝑃𝑇 = 𝐴.
Let 𝐴 be orthogonally diagonalisable, i.e. 𝑃𝐷𝑃𝑇 = 𝐴.
∴ 𝐴𝑇 = (𝑃𝐷𝑃𝑇 )𝑇 = (𝑃𝑇 )𝑇 𝐷𝑇 𝑃𝑇 = 𝑃𝐷𝑃𝑇 = 𝐴
Dus is 𝐴 simmetries. / Thus 𝐴 is symmetric.
(Bewys van / Proof of (⟸) word gelaat. / is omitted.)
7
 Definisie / Definition
Die versameling van eiewaardes van 'n matriks
𝐴, word die spektrum van 𝐴 genoem.

The set of eigenvalues of a matrix 𝐴 , is called

8
 Die Spektraalstelling vir simmetriese matrikse (St 3)
The Spectral Theorem for symmetric matrices (Thm 3)
'n Simmetriese 𝑛 × 𝑛 matriks 𝐴 het die vlg. eienskappe:
A symmetric 𝑛 × 𝑛 matrix 𝐴 has the following properties:
a. 𝐴 het 𝑛 reële eiewaardes, as mens multiplisiteite intel.
𝐴 has 𝑛 real eigenvalues, counting multiplicities.
b. dim𝐸𝜆 = alg. mult(𝜆) ∀ 𝜆 in spektrum / in spectrum A.
c. Die eieruimtes is onderling ortogonaal, dws eievektore
wat hoort by verskillende eiewaardes is ortogonaal.
The eigenspaces are mutually orthogonal, i.e.
eigenvectors corresponding to different eigenvalues are
orthogonal.
d. 𝐴 is ortogonaal diagonaliseerbaar
𝐴 is orthogonally diagonalizable.
9
Spektraalontbinding / Spectral decomposition
Gestel 𝐴 = 𝑃𝐷𝑃−1 , waar die kolomme van 𝑃
ortonormale eievektore 𝒖1, … , 𝒖𝑛 van 𝐴 is en die
ooreenstemmende eiewaardes 𝜆1, … , 𝜆𝑛 in die
diagonaalmatriks 𝐷 voorkom.
Suppose A  PDP 1, where the columns of 𝑃 are
orthonormal eigenvectors 𝒖1, … , 𝒖𝑛 of 𝐴 and the
corresponding eigenvalues 𝜆1, … , 𝜆𝑛 are in the diagonal
matrix 𝐷.
Dan volg, omdat / Then, since
P 1  PT

10
Spektraalontbinding 2 / Spectral decomposition2

 λ1 0   u1T 
A  PDPT   u1 un   
  
 0 λ n   u Tn 
 u1T 
 
  λ1u1 λnu n   
 u Tn 

11
Spektraalontbinding 3 / Spectral decomposition3
Deur die kolom-ryuitbreiding van 'n produk te gebruik, volg
Using the column-row expansion of a product, we can
write
A  λ1u1u1T  λ 2 u 2 u T2   λ n u n u Tn*
Ons noem dit 'n spektraalontbinding van 𝐴.
We call this a spectral decomposition of A.
O Elke term van * is 'n 𝑛 × 𝑛 matriks van rang 1.
O Each term in * is an 𝑛 × 𝑛 matrix of rank 1.
O Each matrix 𝒖𝑗 𝒖𝑗 𝑇 is a projection matrix in the sense
that for each 𝒙 in ℝ𝑛 , the vector 𝒖𝑗 𝒖𝑗 𝑇 𝒙 is the
orthogonal projection of 𝒙 onto the subspace spanned
by 𝒖𝑗.
12