Eigenvalu=1 always true in Markov Matrix
Intuitive Explanation
Is that sure that as long as a matrix whose sum of rows equal to 1s, then the matrix has eigenvalue 1?
So that suppose
v
⃗
′
=
v
⃗
T
P
\vec{v}' = \vec{v}^TP
v′=vTP and
v
⃗
\vec{v}
v is a eigenvector so that
v
⃗
′
=
λ
v
⃗
\vec{v}'= \lambda\vec{v}
v′=λv. Because
v
⃗
′
\vec{v}'
v′ is also a probability distribution so sum of the elements in
v
⃗
′
\vec{v}'
v′ is also equal to one. Therefore, only
λ
=
1
\lambda=1
λ=1 will satisfy the above condition, otherwise sum will be not equal to one.
Math Explanation
To find the eigenvalues of the following Markov matrix P
v
⃗
T
P
=
λ
v
⃗
T
(
∗
∗
∗
∗
)
\vec{v}^T P = \lambda\vec{v}^T (****)
vTP=λvT(∗∗∗∗),
We can find the alternative eigenvalues of
P
T
P^T
PT in (**):
v
⃗
P
T
=
λ
v
⃗
(
∗
∗
)
\vec{v} P^T = \lambda\vec{v} (**)
vPT=λv(∗∗)
Since d e t ( P − λ I ) = 0 det(P-\lambda I)=0 det(P−λI)=0, is equivalent to d e t ( P T − λ I ) det(P^T-\lambda I) det(PT−λI), the eigenvalues are also same in two settings (****) (**).
The fact is that the sum of rows for each column in
P
T
P^T
PT are 1s, then
[
1
,
1
,
.
.
.
,
1
]
P
T
=
[
1
,
1
,
.
.
.
,
1
]
[1,1,...,1] P^T = [1,1,...,1]
[1,1,...,1]PT=[1,1,...,1].
Hence, the eigenvalue=1 will always can be obtained in (**) by the above fact, which is also shared in (). So we can plug
λ
=
1
\lambda=1
λ=1 back in () to get its corresponding eigenvectors.
Eigenvalues <=1 Always True
Prove by contradition
Suppose
λ
>
1
\lambda>1
λ>1
Suppose after n time steps, we have
v
⃗
T
P
n
=
λ
n
v
⃗
T
\vec{v}^TP^n = \lambda^n \vec{v}^T
vTPn=λnvT.
Then RHS contain elements greater than 1
So some elements in
[
P
n
]
i
j
>
1
[P^n]_{ij} >1
[Pn]ij>1, which contradicts the fact that all the elements in
P
n
<
=
1
P^n <=1
Pn<=1.
Therefore, the assumption of
λ
>
1
\lambda>1
λ>1 is wrong
So
λ
\lambda
λ needs to be less than or euqual to 1,
λ
<
=
1
\lambda<=1
λ<=1