Chapter 3 紧算子与Fredholm算子
紧算子:
X
,
Y
\mathscr{X}, \mathscr{Y}
X,Y
B
B
B 空间,
A
:
X
→
Y
A: \mathscr{X} \rightarrow \mathscr{Y}
A:X→Y 线性. 称
A
A
A 是紧算子, 如果
A
(
B
1
)
‾
\overline{A\left(B_{1}\right)}
A(B1) 在
Y
\mathscr{Y}
Y 中是紧集, 其中
B
1
B_{1}
B1 是
X
\mathscr{X}
X 中的单位球.
一切紧算子的集合记作
C
(
X
,
Y
)
\mathfrak{C}(\mathscr{X}, \mathscr{Y})
C(X,Y), 当
X
=
Y
\mathscr{X}=\mathscr{Y}
X=Y 时, 记作
C
(
X
)
\mathfrak{C}(\mathscr{X})
C(X)
注:
A
A
A紧
⇔
\Leftrightarrow
⇔对于
X
\mathscr{X}
X 中的任意有界集
B
,
A
(
B
)
‾
B, \overline{A(B)}
B,A(B) 在
Y
\mathscr{Y}
Y 中是紧集
⇔
\Leftrightarrow
⇔对任意有界点列
{
x
n
}
⊂
X
,
{
A
x
n
}
\left\{x_{n}\right\} \subset \mathscr{X},\left\{A x_{n}\right\}
{xn}⊂X,{Axn} 有收敛子列(紧等价于自列紧)
性质:
(1)
C
(
X
,
Y
)
⊂
L
(
X
,
Y
)
\mathfrak{C}(\mathscr{X}, \mathscr{Y}) \subset \mathscr{L}(\mathscr{X}, \mathscr{Y})
C(X,Y)⊂L(X,Y)且闭
(2)
C
(
X
,
Y
)
\mathfrak{C}(\mathscr{X}, \mathscr{Y})
C(X,Y)是线性空间
(3) 设
A
∈
C
(
X
,
Y
)
A \in \mathfrak{C}(\mathscr{X}, \mathscr{Y})
A∈C(X,Y), 又设
X
0
⊂
X
\mathscr{X}_{0} \subset \mathscr{X}
X0⊂X 是一个闭线性子空间, 那么
A
0
≜
A
∣
X
0
∈
C
(
X
0
,
Y
)
A_{0} \triangleq A \mid \mathscr{X}_{0} \in \mathfrak{C}\left(\mathscr{X}_{0}, \mathscr{Y}\right)
A0≜A∣X0∈C(X0,Y)
(4) 若
A
∈
C
(
X
,
Y
)
A \in \mathfrak{C}(\mathscr{X}, \mathscr{Y})
A∈C(X,Y), 则
R
(
A
)
R(A)
R(A) 可分
(5) 若
A
∈
L
(
X
,
Y
)
A \in \mathscr{L}(\mathscr{X}, \mathscr{Y})
A∈L(X,Y), 而
B
∈
L
(
Y
,
Z
)
B \in \mathscr{L}(\mathscr{Y}, \mathscr{Z})
B∈L(Y,Z), 并且这两个算子中 有一个是紧的, 则
B
A
∈
C
(
X
,
Z
)
B A \in \mathfrak{C}(\mathscr{X}, \mathscr{Z})
BA∈C(X,Z).
全连续: x n ⇀ x ⟹ A x n → A x x_{n} \rightharpoonup x \Longrightarrow A x_{n} \rightarrow A x xn⇀x⟹Axn→Ax
命题:紧算子必定全连续,自反空间中的全连续算子必定紧
定理: T ∈ C ( X , Y ) ⟺ T ∗ ∈ C ( Y ∗ , X ∗ ) T \in \mathfrak{C}(\mathscr{X}, \mathscr{Y}) \Longleftrightarrow T^{*} \in \mathbb{C}\left(\mathscr{Y}^*, \mathscr{X}^{*}\right) T∈C(X,Y)⟺T∗∈C(Y∗,X∗)
紧算子的构造:
有穷秩算子: 设 T ∈ L ( X , Y ) T \in \mathscr{L}(\mathscr{X}, \mathscr{Y}) T∈L(X,Y), 若 dim R ( T ) < ∞ \operatorname{dim} R(T)<\infty dimR(T)<∞,一切有穷秩算子的集合记作 F ( X , Y ) F(\mathscr{X}, \mathscr{Y}) F(X,Y)(显然, F ⊂ C F\subset C F⊂C)
秩1算子:设 f ∈ X ∗ , y ∈ Y f \in \mathscr{X}^{*}, y \in \mathscr{Y} f∈X∗,y∈Y, 用 y ⊗ f y \otimes f y⊗f 表示下列算子: x ↦ ⟨ f , x ⟩ y ( ∀ x ∈ X ) x \mapsto\langle f, x\rangle y \quad(\forall x \in \mathscr{X}) x↦⟨f,x⟩y(∀x∈X),称为秩1算子
有穷秩算子的分解: T ∈ F ( X , Y ) T \in F(\mathscr{X}, \mathscr{Y}) T∈F(X,Y) ⇔ \Leftrightarrow ⇔ ∃ y i ∈ Y \exists y_{i} \in \mathscr{Y} ∃yi∈Y 以 及 f i ∈ X ∗ ( i = 1 , 2 , ⋯ , n ) f_{i} \in \mathscr{X}^{*}(i=1,2, \cdots, n) fi∈X∗(i=1,2,⋯,n), 使得 T = ∑ i = 1 n y i ⊗ f i T=\sum_{i=1}^{n} y_{i} \otimes f_{i} T=∑i=1nyi⊗fi
有穷秩算子的逼近:
(1) 在Hilbert空间上, F ( X , Y ) ‾ = C ( X , Y ) \overline{F(\mathscr{X}, \mathscr{Y})}=\mathfrak{C}(\mathscr{X}, \mathscr{Y}) F(X,Y)=C(X,Y)(思路:在有穷 ϵ \epsilon ϵ-网上做正交投影)
(2) **在Banach空间上,如果存在一组Schauder基:
{
e
n
}
n
=
1
∞
⊂
X
\left\{e_{n}\right\}_{n=1}^{\infty} \subset \mathscr{X}
{en}n=1∞⊂X 为
X
\mathscr{X}
X:
∀
x
∈
X
\forall x \in \mathscr{X}
∀x∈X, 存在唯一的一个序列
{
C
n
(
x
)
}
\left\{C_{n}(x)\right\}
{Cn(x)}, 使得
x
=
lim
N
→
∞
∑
n
=
1
N
C
n
(
x
)
e
n
x=\lim _{N \rightarrow \infty} \sum_{n=1}^{N} C_{n}(x) e_{n}
x=limN→∞∑n=1NCn(x)en ,则也有
F
(
X
,
Y
)
‾
=
C
(
X
,
Y
)
\overline{F(\mathscr{X}, \mathscr{Y})}=\mathfrak{C}(\mathscr{X}, \mathscr{Y})
F(X,Y)=C(X,Y),但是Schauder基并不一定存在
Reisz-Fredholm理论:
需求:求解微分方程(形如: x ( t ) = ∫ 0 1 K ( t , s ) x ( s ) d s + y ( t ) x(t)=\int_0^1 K(t,s)x(s)ds+y(t) x(t)=∫01K(t,s)x(s)ds+y(t)),其定义了一个算子 T x = y , T = I − A , Tx=y, T=I-A, Tx=y,T=I−A,且 T A TA TA紧
Fredholm二择一:关于该方程仅存在两种可能:(1) ∀ y ∈ L 2 [ 0 , 1 ] \forall y\in L^2[0,1] ∀y∈L2[0,1],方程存在唯一解 (2) y = θ y=\theta y=θ时,方程有非零解
对任意的 M ⊂ X , N ⊂ X ∗ M \subset \mathscr{X}, N \subset \mathscr{X}^{*} M⊂X,N⊂X∗, 记 ⊥ M ≜ { f ∈ X ∗ ∣ ⟨ f , x ⟩ = 0 , ∀ x ∈ M } , N ⊥ ≜ { x ∈ X ∣ ⟨ f , x ⟩ = 0 , ∀ f ∈ N } . \begin{gathered} { }^{\perp} M \triangleq\left\{f \in \mathscr{X}^{*} \mid\langle f, x\rangle=0, \forall x \in M\right\}, \\ N^{\perp} \triangleq\{x \in \mathscr{X} \mid\langle f, x\rangle=0, \forall f \in N\} . \end{gathered} ⊥M≜{f∈X∗∣⟨f,x⟩=0,∀x∈M},N⊥≜{x∈X∣⟨f,x⟩=0,∀f∈N}.
又若 f ∈ X ∗ , x ∈ X f \in \mathscr{X}^{*}, x \in \mathscr{X} f∈X∗,x∈X, 满足 ⟨ f , x ⟩ = 0 \langle f, x\rangle=0 ⟨f,x⟩=0, 便简单地记作 f ⊥ x f \perp x f⊥x
则:Fredholm理论: T = I − A , A T=I-A,A T=I−A,A紧=> (1) N ( T ) = { θ } ⟹ R ( T ) = X N(T)=\{\theta\} \Longrightarrow R(T)=\mathscr{X} N(T)={θ}⟹R(T)=X (2) σ ( T ) = σ ( T ∗ ) ‾ \sigma(T)=\overline{\sigma\left(T^{*}\right)} σ(T)=σ(T∗)(共轭),且 dim N ( T ) = dim N ( T ∗ ) < ∞ \operatorname{dim} N(T)=\operatorname{dim} N\left(T^{*}\right)<\infty dimN(T)=dimN(T∗)<∞ (3) R ( T ) = N ( T ∗ ) ⊥ , R ( T ∗ ) = ⊥ N ( T ) R(T)=N\left(T^{*}\right)^{\perp}, R\left(T^{*}\right)={ }^{\perp} N(T) R(T)=N(T∗)⊥,R(T∗)=⊥N(T)
余维数: codim ( M ) = dim ( X / M ) \operatorname{codim} (M)=\dim(\mathscr{X}/M) codim(M)=dim(X/M)
紧算子的谱理论:
紧算子的谱分布:若 A ∈ C ( X ) A \in \mathfrak{C}(\mathscr{X}) A∈C(X), 则:(1) 0 ∈ σ ( A ) 0 \in \sigma(A) 0∈σ(A), 除非 dim X < ∞ \operatorname{dim} \mathscr{X}<\infty dimX<∞ (2) σ ( A ) \ { 0 } = σ p ( A ) \ { 0 } \sigma(A) \backslash\{0\}=\sigma_{p}(A) \backslash\{0\} σ(A)\{0}=σp(A)\{0} (3) σ p ( A ) \sigma_{p}(A) σp(A) 至多以 0 为聚点.
翻译:在无穷维空间上,只有三种可能:(1) σ ( A ) = { 0 } \sigma(A)=\{0\} σ(A)={0} (2) σ ( A ) = { 0 , λ 1 , λ 2 , ⋯ , λ n } \sigma(A)=\left\{0, \lambda_{1}, \lambda_{2}, \cdots, \lambda_{n}\right\} σ(A)={0,λ1,λ2,⋯,λn} (3) σ ( A ) = { λ 1 , λ 2 , ⋯ , λ n , ⋯ } \sigma(A)=\left\{\lambda_{1}, \lambda_{2}, \cdots, \lambda_{n}, \cdots\right\} σ(A)={λ1,λ2,⋯,λn,⋯}, 其中 λ n → 0 \lambda_{n} \rightarrow 0 λn→0
不变子空间:设 X \mathscr{X} X 是一个 B B B 空间, M ⊂ X M \subset \mathscr{X} M⊂X ,若 A ( M ) ⊂ M A(M) \subset M A(M)⊂M则称为不变子空间
常见不变子空间:(1)平凡不变子空间: { θ } , X \{\theta\},\mathscr{X} {θ},X (2) λ ∈ σ p ( A ) \lambda \in \sigma_{p}(A) λ∈σp(A), 即 λ \lambda λ 是 A A A 的特征值, 则 N ( λ I − A ) N(\lambda I-A) N(λI−A) 为不变子空间 (3) ∀ y ∈ X \forall y \in \mathscr{X} ∀y∈X, 若记 L y ≜ { P ( A ) y ∣ P L_{y} \triangleq\{P(A) y \mid P Ly≜{P(A)y∣P 是任意多项式 } \} }, 则 L y L_{y} Ly 是 A A A的不变子空间.
定理:若 dim X ⩾ 2 \operatorname{dim} \mathscr{X} \geqslant 2 dimX⩾2, 则 ∀ A ∈ C ( X ) , A \forall A \in \mathfrak{C}(\mathscr{X}), A ∀A∈C(X),A 必有非平凡的闭不变子空间.
Fredholm算子: 设
X
,
Y
\mathscr{X}, \mathscr{Y}
X,Y 是 Banach 空间,
T
∈
L
(
X
,
Y
)
T \in \mathscr{L}(\mathscr{X}, \mathscr{Y})
T∈L(X,Y) 称为一个 Fredholm 算子, 是指:
(1)
R
(
T
)
R(T)
R(T) 是闭的;
(2)
dim
N
(
T
)
<
∞
\operatorname{dim} N(T)<\infty
dimN(T)<∞;
(3)
codim
R
(
T
)
<
∞
\operatorname{codim} R(T)<\infty
codimR(T)<∞
X
→
Y
\mathscr{X} \rightarrow \mathscr{Y}
X→Y 的一切 Fredholm 算子的全体记作
F
(
X
,
Y
)
\mathscr{F}(\mathscr{X}, \mathscr{Y})
F(X,Y), 特别地, 当
Y
=
X
\mathscr{Y}=\mathscr{X}
Y=X 时, 记作
F
(
X
)
\mathscr{F}(\mathscr{X})
F(X).
指标: ind ( T ) = dim N ( T ) − codim R ( T ) \operatorname{ind}(T) =\dim N(T)-\operatorname{codim} R(T) ind(T)=dimN(T)−codimR(T),从而 I − C , C I-C,C I−C,C紧为一个Fredholm算子
则:左右移位算子的指标分别为正负1
Fredholm算子的结构:
(1) 若
T
∈
F
(
X
,
Y
)
T \in \mathscr{F}(\mathscr{X}, \mathscr{Y})
T∈F(X,Y), 则必有
S
∈
L
(
Y
,
X
)
S \in \mathscr{L}(\mathscr{Y}, \mathscr{X})
S∈L(Y,X) 以 及
A
1
∈
C
(
X
)
,
A
2
∈
C
(
Y
)
A_{1} \in \mathfrak{C}(\mathscr{X}), A_{2} \in \mathfrak{C}(\mathscr{Y})
A1∈C(X),A2∈C(Y), 使得
S
T
=
I
x
−
A
1
,
T
S
=
I
y
−
A
2
S T=I_{x}-A_{1}, T S=I_{y}-A_{2}
ST=Ix−A1,TS=Iy−A2,其中
I
x
,
I
y
I_{x}, I_{y}
Ix,Iy 分别表示
X
\mathscr{X}
X 和
Y
\mathscr{Y}
Y 上的恒同算子.
(2) 如果
T
∈
L
(
X
,
Y
)
T \in \mathscr{L}(\mathscr{X}, \mathscr{Y})
T∈L(X,Y), 又有
R
1
,
R
2
∈
L
(
Y
,
X
)
R_{1}, R_{2} \in \mathscr{L}(\mathscr{Y}, \mathscr{X})
R1,R2∈L(Y,X) 以及
A
1
∈
A_{1} \in
A1∈
C
(
X
)
,
A
2
∈
C
(
Y
)
\mathfrak{C}(\mathscr{X}), A_{2} \in \mathfrak{C}(\mathscr{Y})
C(X),A2∈C(Y), 使得
R
1
T
=
I
x
−
A
1
,
T
R
2
=
I
y
−
A
2
R_{1} T=I_{x}-A_{1}, T R_{2}=I_{y}-A_{2}
R1T=Ix−A1,TR2=Iy−A2,则
T
∈
F
(
X
,
Y
)
T \in \mathscr{F}(\mathscr{X}, \mathscr{Y})
T∈F(X,Y)
R 1 , R 2 R_1,R_2 R1,R2分别称为 T T T的左右正则化子,意义为:Fredholm算子在左右正则化子的作用下确实与 I − 紧 算 子 I-紧算子 I−紧算子相同
定理:若 T 1 ∈ F ( X , Y ) , T 2 ∈ F ( Y , Z ) T_{1} \in \mathscr{F}(\mathscr{X}, \mathscr{Y}), T_{2} \in \mathscr{F}(\mathscr{Y}, \mathscr{Z}) T1∈F(X,Y),T2∈F(Y,Z), 其中 X , Y \mathscr{X}, \mathscr{Y} X,Y, Z \mathscr{Z} Z 都是 Banach 空间, 则 T 2 T 1 ∈ F ( X , Z ) T_{2} T_{1} \in \mathscr{F}(\mathscr{X}, \mathscr{Z}) T2T1∈F(X,Z), 且 ind ( T 2 T 1 ) = ind ( T 1 ) + ind ( T 2 ) \operatorname{ind}\left(T_{2} T_{1}\right)=\operatorname{ind}\left(T_{1}\right)+\operatorname{ind}\left(T_{2}\right) ind(T2T1)=ind(T1)+ind(T2)
线性微扰稳定性:若 T ∈ F ( X , Y ) T \in \mathscr{F}(\mathscr{X}, \mathscr{Y}) T∈F(X,Y), 则存在 ε > 0 \varepsilon>0 ε>0, 使得当 S ∈ S \in S∈ L ( X , Y ) \mathscr{L}(\mathscr{X}, \mathscr{Y}) L(X,Y), 且 ∥ S ∥ < ε \|S\|<\varepsilon ∥S∥<ε 时, 有 T + S ∈ F ( X , Y ) T+S \in \mathscr{F}(\mathscr{X}, \mathscr{Y}) T+S∈F(X,Y),并且 ind ( T + S ) = ind ( T ) \operatorname{ind}(T+S)=\operatorname{ind}(T) ind(T+S)=ind(T)