Try   HackMD

Find the spectral decomposition

Problem

Let

A=[0011001111001100].

Find the spectral decomposition of

A.

Thought

The spectral decomposition of

A is a way to write
A
as

A=j=1qμjPj

such that

  • Pi2=Pi
    for any
    i
    ,
  • PiPj=O
    for any
    i
    and
    j
    , and
  • j=1qPj=In
    .

Intuitively,

Pj's are projection matrices whose ranges are orthogonal to each other and span the whole space.

In practice, we may first find the eigenvalues

λ1,,λn and an orthonormal basis
{u1,,un}
of
Rn
. Then

A=λ1u1u1++λnunun=μ1P1++μqPq,

where

μ1,,μq are the distinct eigenvalues of
A
and
Pj
is the sum of
uiui
for all
λi=μj
.

Sample answer

According to Find an orthonormal basis composed of eigenvectors, we already have

λ1=2,
λ2=2
,
λ3=λ4=0
. Thus, we may set
μ1=2
,
μ2=2
and
μ3=0
with
q=3
.

The corresponding eigenvectors are

u1,
u2
,
u3
, and
u4
. As
u3
and
u4
both correspond to the eigenvalue
μ3=0
, we will add them together to form a projection matrix onto
ker(A0I)
. In summary, we have

A=λ1u1u1+λ2u2u2+λ3u3u3+λ4u4u4=μ1u1u1+μ2u2u2+λ3(u3u3+u4u4)=2P1+(2)P2+0P3,

where

P1=14[1111111111111111], P2=14[1111111111111111],

and

P3=12[1100110000110011],

which is the spectral decomposition of

A.

This note can be found at Course website > Learning resources.