Definition:
We call P ( λ ) ∈ M n ( F [ λ ] ) P(\lambda)\in M_n(F[\lambda]) P(λ)∈Mn(F[λ]) a λ \lambda λ-martix.
Attention! P ∈ M n [ F ] P\in M_n[F] P∈Mn[F]!
rank ( A ( λ ) ) : = max { the order of a non-zero minor of A ( λ ) } \text{rank}(A(\lambda)):=\max\{\text{the order of a non-zero minor of }A(\lambda)\} rank(A(λ)):=max{the order of a non-zero minor of A(λ)}.
A ( λ ) is invertible : = ∃ B ( λ ) , s . t . A ( λ ) B ( λ ) = B ( λ ) A ( λ ) = I A(\lambda)\text{ is invertible}:=\exist B(\lambda),\ s.t.\ A(\lambda)B(\lambda)=B(\lambda)A(\lambda)=I A(λ) is invertible:=∃B(λ), s.t. A(λ)B(λ)=B(λ)A(λ)=I.
A ( λ ) , B ( λ ) are equivalent : = ∃ U ( λ ) , V ( λ ) s . t . U ( λ ) A ( λ ) V ( λ ) = B ( λ ) A(\lambda),B(\lambda)\text{ are equivalent}:=\exist U(\lambda),V(\lambda)\ s.t.\ U(\lambda)A(\lambda)V(\lambda)=B(\lambda) A(λ),B(λ) are equivalent:=∃U(λ),V(λ) s.t. U(λ)A(λ)V(λ)=B(λ), where U ( λ ) , V ( λ ) U(\lambda),V(\lambda) U(λ),V(λ) are invertible.
Properties:
A ( λ ) A(\lambda) A(λ) is invertible. ⇔ ∣ A ( λ ) ∣ ∈ U ( F ) . ⇔ A ( λ ) = ∏ P ( λ ) \Leftrightarrow|A(\lambda)|\in U(F).\Leftrightarrow A(\lambda)=\prod P(\lambda) ⇔∣A(λ)∣∈U(F).⇔A(λ)=∏P(λ), where P ( λ ) P(\lambda) P(λ) are elementary λ \lambda λ-matrices.
As in here, the second condition is ∣ A ( λ ) ∣ = c ≠ 0 |A(\lambda)|=c\neq0 ∣A(λ)∣=c=0.
Attention! That the result is not zero is not equivalent with the condition above! It can be other polynomials with positive degree.
Elementary Transformation
Similarly you can find out matrices betokening elementary transformations.
A row/column should be multiplied by a unit, not a zero divisor, because that allows no way back.
λ I − A \lambda I-A λI−A is called the characteristic matrix of A A A.
Obviously, D n D_n Dn of λ I − A \lambda I-A λI−A is the characteristic polynomial of A A A.
Equivalence
A ∼ B . ⇔ λ I − A ≃ λ I − B . A\sim B.\Leftrightarrow\lambda I-A\simeq\lambda I-B. A∼B.⇔λI−A≃λI−B.
This enables you to perform elementary transformations to your heart’s content.
Proof:
1, If there exists number matrices P , Q , s . t . P ( λ I − A ) Q = λ I − B P,Q,\ s.t.\ P(\lambda I-A)Q=\lambda I-B P,Q, s.t. P(λI−A)Q=λI−B.
λ P Q − P A Q = λ I − B \lambda PQ-PAQ=\lambda I-B λPQ−PAQ=λI−B.
(Such matrices can be regarded as polynomials whose coefficients are number matrices.)
Considering its leading coefficient, P Q = I PQ=I PQ=I.
2, It’s not a commutative ring.
U = P ( λ I − A ) + U 0 , V = ( λ I − A ) Q + V 0 U=P(\lambda I-A)+U_0,V=(\lambda I-A)Q+V_0 U=P(λI−A)+U0,V=(λI−A)Q+V0.
3,
U
(
λ
)
(
λ
I
−
A
)
=
(
λ
I
−
B
)
V
−
1
(
λ
)
.
U
(
λ
)
=
(
λ
I
−
B
)
Q
(
λ
)
+
U
0
.
⇒
(
λ
I
−
B
)
Q
(
λ
)
(
λ
I
−
A
)
+
U
0
(
λ
I
−
A
)
=
(
λ
I
−
B
)
V
−
1
(
λ
)
.
⇒
(
λ
I
−
B
)
[
−
Q
(
λ
)
(
λ
I
−
A
)
+
V
−
1
(
λ
)
]
=
U
0
(
λ
I
−
A
)
.
Consider degrees in both side.
⇒
V
−
1
(
λ
)
−
Q
(
λ
)
(
λ
I
−
A
)
∈
M
n
(
F
)
, set down as
P
.
⇒
U
0
(
λ
I
−
A
)
=
(
λ
I
−
B
)
P
.
⇒
P
=
U
0
.
\begin{aligned} & U(\lambda)(\lambda I-A)=(\lambda I-B) V^{-1}(\lambda) . \\ & U(\lambda)=(\lambda I-B) Q(\lambda)+U_{0} . \\ \Rightarrow &(\lambda I-B) Q(\lambda)(\lambda I-A)+U_{0}(\lambda I-A)=(\lambda I-B) V^{-1}(\lambda) . \\ \Rightarrow &(\lambda I-B)\left[-Q(\lambda)(\lambda I-A)+V^{-1}(\lambda)\right]=U_{0}(\lambda I-A) . \\ \text {Consider degrees in both side. } \\ \Rightarrow & V^{-1}(\lambda)-Q(\lambda)(\lambda I-A) \in M_{n}(F) \text {, set down as } P . \\ \Rightarrow & U_{0}(\lambda I-A)=(\lambda I-B) P.\Rightarrow P=U_{0}. \end{aligned}
⇒⇒Consider degrees in both side. ⇒⇒U(λ)(λI−A)=(λI−B)V−1(λ).U(λ)=(λI−B)Q(λ)+U0.(λI−B)Q(λ)(λI−A)+U0(λI−A)=(λI−B)V−1(λ).(λI−B)[−Q(λ)(λI−A)+V−1(λ)]=U0(λI−A).V−1(λ)−Q(λ)(λI−A)∈Mn(F), set down as P.U0(λI−A)=(λI−B)P.⇒P=U0.
Now we prove
P
−
1
∃
P^{-1} \exists
P−1∃.
P
V
(
λ
)
=
I
−
Q
(
λ
)
[
(
λ
I
−
A
)
V
(
λ
)
]
=
I
−
Q
(
λ
)
U
−
1
(
λ
)
(
λ
I
−
B
)
.
Let
V
(
λ
)
=
R
(
λ
)
(
λ
I
−
B
)
+
V
0
.
⇒
P
V
(
λ
)
=
P
R
(
λ
)
(
λ
I
−
B
)
+
P
V
0
.
⇒
(
P
R
(
λ
)
+
Q
(
λ
)
U
−
1
(
λ
)
)
(
λ
I
−
B
)
=
I
−
P
V
0
.
\begin{aligned} & P V(\lambda)=I-Q(\lambda)[(\lambda I-A) V(\lambda)]=I-Q(\lambda) U^{-1}(\lambda)(\lambda I-B) . \\ \text { Let } V(\lambda)=R(\lambda)(\lambda I-B)+V_{0} . \\ \Rightarrow & P V(\lambda)=P R(\lambda)(\lambda I-B)+P V_{0}.\\ \Rightarrow &\left(P R(\lambda)+Q(\lambda) U^{-1}(\lambda)\right)(\lambda I-B)=I-P V_{0} . \end{aligned}
Let V(λ)=R(λ)(λI−B)+V0.⇒⇒PV(λ)=I−Q(λ)[(λI−A)V(λ)]=I−Q(λ)U−1(λ)(λI−B).PV(λ)=PR(λ)(λI−B)+PV0.(PR(λ)+Q(λ)U−1(λ))(λI−B)=I−PV0.
Consider degrees in both side.
⇒
P
R
(
λ
)
+
Q
(
λ
)
U
−
1
(
λ
)
=
0.
⇒
P
V
0
=
I
.
⇒
P
−
1
=
V
0
.
\Rightarrow P R(\lambda)+Q(\lambda) U^{-1}(\lambda)=0 . \Rightarrow P V_{0}=I. \Rightarrow P^{-1}=V_{0}.
⇒PR(λ)+Q(λ)U−1(λ)=0.⇒PV0=I.⇒P−1=V0.
∀ A ( λ ) ∼ B ( λ ) = ( b i , j ( λ ) ) , deg b 1 , 1 ( λ ) = min { deg b i , j } & b 1 , 1 ∣ b i , j \forall A(\lambda)\sim B(\lambda)=(b_{i,j}(\lambda)),\deg b_{1,1}(\lambda) =\min\{\deg b_{i,j}\}\ \&\ b_{1,1}|b_{i,j} ∀A(λ)∼B(λ)=(bi,j(λ)),degb1,1(λ)=min{degbi,j} & b1,1∣bi,j.
Perform division with remainder with element 1,1.
If can not be divided, change the remainder to 1,1.
Finally you get a 1,1 element which can divide all other elements.
Break Bricks in row 1 and column 1.
Repeat such process in lower right corner.
Smith Canonical Form and Invariant Factor
Any
λ
−
\lambda-
λ−matrix is equivalent to such Smith canonical form:
(
d
1
(
x
)
d
2
(
x
)
⋱
)
\left( \begin{matrix} d_1(x)\\ &d_2(x)\\ &&\ddots \end{matrix} \right)
⎝⎛d1(x)d2(x)⋱⎠⎞
d
i
d_i
di is called an invariant factor of
A
(
λ
)
A(\lambda)
A(λ).
d i d_i di has leading coefficient 1 and d i ∣ d i + 1 d_i|d_{i+1} di∣di+1.
Then we prove d i d_i di is independent with transformations you use.
Determinant Factor
Definition:
Determinant factor D k : = ( { all k -order minors of A ( λ ) } ) D_k:=(\{\text{all }k\text{-order minors of }A(\lambda)\}) Dk:=({all k-order minors of A(λ)}).
We know these are independent with elementary transformations, because those transformations don’t change common divisor.
Then we know the canonical form is unique.
D k = ∏ i = 1 k d i D_k=\prod\limits_{i=1}^kd_i Dk=i=1∏kdi.
So D i o r d i D_i\ or\ d_i Di or di are complete invariants of equivalent λ − \lambda- λ−matrices.
For invertible matrix, D n = 1. ⇒ D i = 1. ⇒ d i = 1 D_n=1.\Rightarrow D_i=1.\Rightarrow d_i=1 Dn=1.⇒Di=1.⇒di=1, then I I I is the canonical form of invertible matrices.
Define λ I − A \lambda I-A λI−A’s invariant factors and elementary factors A A A’s invariant factors and elementary factors.
When determining determinant factors, you start with biggest one, because it can determine former ones, if you are lucky.
Elementary Factor
All irreducible divisors which appear in the decomposition of d i d_i di, are called elementary factors of A ( λ ) A(\lambda) A(λ). If it repeats, set down all.
Align them in ascending order of
q
q
q: (Less than
n
n
n? Fill with
1
1
1.)
(
x
+
p
1
)
q
1
,
1
(
x
+
p
1
)
q
1
,
2
⋯
(
x
+
p
2
)
q
2
,
1
(
x
+
p
2
)
q
2
,
2
⋯
\begin{matrix} (x+p_1)^{q_{1,1}}&(x+p_1)^{q_{1,2}}&\cdots\\ (x+p_2)^{q_{2,1}}&(x+p_2)^{q_{2,2}}&\cdots\\ \end{matrix}
(x+p1)q1,1(x+p2)q2,1(x+p1)q1,2(x+p2)q2,2⋯⋯
Attention: This is just an example in
C
\mathbb C
C! Irreducible divisors can be very different in
Q
Q
Q or other sets.
You can figure that the i t h i^{th} ith column multiplied together makes d i d_i di.
Elementary factors are easier to figure out:
Turn A ( λ ) A(\lambda) A(λ) to a diagonal matrix by elementary transformations; then uniquely decompose all diagonal elements, elementary factors got.
Proof:
Just need to prove:
If
(
f
i
,
g
j
)
=
1
,
∀
i
,
j
∈
{
0
,
1
}
(f_i,g_j)=1,\forall i,j\in\{0,1\}
(fi,gj)=1,∀i,j∈{0,1}, then:
(
f
1
g
1
0
0
f
2
g
2
)
∼
(
f
2
g
1
0
0
f
1
g
2
)
.
\left( \begin{matrix} f_1g_1&0\\ 0&f_2g_2\\ \end{matrix} \right) \sim \left( \begin{matrix} f_2g_1&0\\ 0&f_1g_2\\ \end{matrix} \right).
(f1g100f2g2)∼(f2g100f1g2).
Obviously, they have same determinant factors.