Sea
A
∈
M
m
×
n
(
K
)
{\displaystyle A\in {\mathcal {M}}_{m\times n}(K)}
,
K
{\displaystyle K}
un cuerpo, una matriz con coeficientes
a
i
j
∈
K
{\displaystyle a_{ij}\in K}
. Se define el espacio columna, el espacio fila y el espacio nulo de
A
{\displaystyle A}
, respectivamente, como
Col
(
A
)
:=
⟨
(
a
11
,
…
,
a
m
1
)
,
…
,
(
a
1
n
,
…
,
a
m
n
)
⟩
⊆
K
m
{\displaystyle \operatorname {Col} (A):=\langle (a_{11},\ldots ,a_{m1}),\ldots ,(a_{1n},\ldots ,a_{mn})\rangle \subseteq K^{m}}
;
Fil
(
A
)
:=
⟨
(
a
11
,
…
,
a
1
n
)
,
…
,
(
a
m
1
,
…
,
a
m
n
)
⟩
⊆
K
n
{\displaystyle \operatorname {Fil} (A):=\langle (a_{11},\ldots ,a_{1n}),\ldots ,(a_{m1},\ldots ,a_{mn})\rangle \subseteq K^{n}}
;
Nul
(
A
)
:=
{
x
∈
K
n
:
A
x
=
0
}
{\displaystyle \operatorname {Nul} (A):=\{x\in K^{n}:Ax=0\}}
;
donde
0
{\displaystyle 0}
denota el vector nulo del espacio vectorial
K
m
{\displaystyle K^{m}}
.
Sea
A
:=
(
−
1
1
0
1
1
−
2
−
2
1
1
)
{\displaystyle A:={\begin{pmatrix}-1&1&0\\1&1&-2\\-2&1&1\end{pmatrix}}}
. Entonces:
Col
(
A
)
=
⟨
(
−
1
,
1
,
−
2
)
,
(
1
,
1
,
1
)
,
(
0
,
−
2
,
1
)
⟩
=
⟨
(
−
1
,
1
,
−
2
)
,
(
1
,
1
,
1
)
⟩
{\displaystyle \operatorname {Col} (A)=\langle (-1,1,-2),(1,1,1),(0,-2,1)\rangle =\langle (-1,1,-2),(1,1,1)\rangle }
;
Fil
(
A
)
=
⟨
(
−
1
,
1
,
0
)
,
(
1
,
1
,
−
2
)
,
(
−
2
,
1
,
1
)
⟩
=
⟨
(
−
1
,
1
,
0
)
,
(
1
,
1
,
−
2
)
⟩
{\displaystyle \operatorname {Fil} (A)=\langle (-1,1,0),(1,1,-2),(-2,1,1)\rangle =\langle (-1,1,0),(1,1,-2)\rangle }
;
Nul
(
A
)
=
⟨
(
1
,
1
,
1
)
⟩
{\displaystyle \operatorname {Nul} (A)=\langle (1,1,1)\rangle }
. La matriz no tiene por qué ser cuadrada; veamos otro ejemplo:
Sea
B
:=
(
−
1
2
3
−
6
−
2
4
)
{\displaystyle B:={\begin{pmatrix}-1&2\\3&-6\\-2&4\end{pmatrix}}}
. Entonces:
Col
(
B
)
=
⟨
(
−
1
,
3
,
−
2
)
,
(
2
,
−
6
,
4
)
⟩
=
⟨
(
−
1
,
3
,
−
2
)
⟩
{\displaystyle \operatorname {Col} (B)=\langle (-1,3,-2),(2,-6,4)\rangle =\langle (-1,3,-2)\rangle }
;
Fil
(
B
)
=
⟨
(
−
1
,
2
)
,
(
3
,
−
6
)
,
(
−
2
,
4
)
⟩
=
⟨
(
−
1
,
2
)
⟩
{\displaystyle \operatorname {Fil} (B)=\langle (-1,2),(3,-6),(-2,4)\rangle =\langle (-1,2)\rangle }
;
Nul
(
B
)
=
⟨
(
2
,
1
)
⟩
{\displaystyle \operatorname {Nul} (B)=\langle (2,1)\rangle }
.
Para las relaciones de ortogonalidades entre conjuntos, siempre se considera el producto escalar estándar de
K
m
{\displaystyle K^{m}}
o
K
n
{\displaystyle K^{n}}
:
Col
(
A
T
)
=
Fil
(
A
)
{\displaystyle \operatorname {Col} (A^{T})=\operatorname {Fil} (A)}
Fil
(
A
T
)
=
Col
(
A
)
{\displaystyle \operatorname {Fil} (A^{T})=\operatorname {Col} (A)}
Nul
(
A
)
⊥
Fil
(
A
)
{\displaystyle \operatorname {Nul} (A)\perp \operatorname {Fil} (A)}
Col
(
A
)
⊥
Nul
(
A
T
)
{\displaystyle \operatorname {Col} (A)\perp \operatorname {Nul} (A^{T})}
dim
(
Col
(
A
)
)
=
dim
(
Col
(
A
T
)
)
=
rg
(
A
)
=
rg
(
A
T
)
{\displaystyle \dim(\operatorname {Col} (A))=\dim(\operatorname {Col} (A^{T}))=\operatorname {rg} (A)=\operatorname {rg} (A^{T})}
.
Si
A
=
(
a
i
j
)
∈
M
n
×
n
(
K
)
{\displaystyle A=(a_{ij})\in {\mathcal {M}}_{n\times n}(K)}
y además las columnas de
A
{\displaystyle A}
,
{
(
a
11
,
…
,
a
n
1
)
,
…
,
(
a
1
n
,
…
,
a
n
n
)
}
{\displaystyle \{(a_{11},\ldots ,a_{n1}),\ldots ,(a_{1n},\ldots ,a_{nn})\}}
, forman un conjunto linealmente independiente de
K
n
{\displaystyle K^{n}}
, entonces
det
(
A
)
≠
0
{\displaystyle \det(A)\neq 0}
, o sea, la matriz es invertible.
Si
A
=
(
a
i
j
)
∈
M
n
×
n
(
K
)
{\displaystyle A=(a_{ij})\in {\mathcal {M}}_{n\times n}(K)}
y además
Nul
(
A
)
≠
{
0
}
{\displaystyle \operatorname {Nul} (A)\neq \{0\}}
, entonces
det
(
A
)
=
0
{\displaystyle \det(A)=0}
, o sea, la matriz no es invertible.
dim
(
Col
(
A
)
)
+
dim
(
Nul
(
A
)
)
=
n
{\displaystyle \dim(\operatorname {Col} (A))+\dim(\operatorname {Nul} (A))=n}
.
Sean
A
∈
M
n
×
m
(
K
)
{\displaystyle A\in {\mathcal {M}}_{n\times m}(K)}
y
B
∈
M
p
×
n
(
K
)
{\displaystyle B\in {\mathcal {M}}_{p\times n}(K)}
. Si
x
∈
Col
(
B
A
)
{\displaystyle x\in \operatorname {Col} (BA)}
, entonces existe
y
∈
K
m
{\displaystyle y\in K^{m}}
tal que
B
A
y
=
x
{\displaystyle BAy=x}
. Si tomamos
w
=
A
y
{\displaystyle w=Ay}
, entonces
B
w
=
x
{\displaystyle Bw=x}
, así que
x
∈
Col
(
B
)
{\displaystyle x\in \operatorname {Col} (B)}
. Por lo tanto,
Col
(
B
A
)
⊆
Col
(
B
)
{\displaystyle \operatorname {Col} (BA)\subseteq \operatorname {Col} (B)}
. Además
Col
(
B
A
)
=
Col
(
B
)
{\displaystyle \operatorname {Col} (BA)=\operatorname {Col} (B)}
si y solo si
rg
(
A
)
=
n
{\displaystyle \operatorname {rg} (A)=n}
.
Sean
A
∈
M
n
×
m
(
K
)
{\displaystyle A\in {\mathcal {M}}_{n\times m}(K)}
y
B
∈
M
p
×
n
(
K
)
{\displaystyle B\in {\mathcal {M}}_{p\times n}(K)}
—en particular,
B
A
∈
M
p
×
m
(
K
)
{\displaystyle BA\in {\mathcal {M}}_{p\times m}(K)}
—. Entonces si
A
x
=
0
{\displaystyle Ax=0}
, también se tiene que
B
A
x
=
0
{\displaystyle BAx=0}
. Así,
Nul
(
A
)
⊆
Nul
(
B
A
)
{\displaystyle \operatorname {Nul} (A)\subseteq \operatorname {Nul} (BA)}
, y ocurre que
Nul
(
A
)
=
Nul
(
B
A
)
{\displaystyle \operatorname {Nul} (A)=\operatorname {Nul} (BA)}
si y solo si
rg
(
B
)
=
n
{\displaystyle \operatorname {rg} (B)=n}
.
Supongamos que
K
=
R
{\displaystyle K=\mathbb {R} }
y sea
A
∈
M
m
×
n
(
R
)
{\displaystyle A\in {\mathcal {M}}_{m\times n}(\mathbb {R} )}
. Veamos que
Nul
(
A
)
=
Nul
(
A
T
A
)
{\displaystyle \operatorname {Nul} (A)=\operatorname {Nul} (A^{T}A)}
. Sea
x
∈
Nul
(
A
)
{\displaystyle x\in \operatorname {Nul} (A)}
, entonces
A
x
=
0
{\displaystyle Ax=0}
, por lo que
A
T
A
x
=
0
{\displaystyle A^{T}Ax=0}
. Por otro lado, si
x
∈
Nul
(
A
T
A
)
{\displaystyle x\in \operatorname {Nul} (A^{T}A)}
, tenemos que
A
T
A
x
=
0
{\displaystyle A^{T}Ax=0}
, por lo tanto
x
T
A
T
A
x
=
0
{\displaystyle x^{T}A^{T}Ax=0}
. Como
x
T
A
T
A
x
=
(
A
x
)
T
A
x
=
⟨
A
x
,
A
x
⟩
{\displaystyle x^{T}A^{T}Ax=(Ax)^{T}Ax=\langle Ax,Ax\rangle }
, donde
⟨
⋅
,
⋅
⟩
{\displaystyle \langle \cdot ,\cdot \rangle }
denota el producto escalar estándar de
R
m
{\displaystyle \mathbb {R} ^{m}}
, necesariamente
A
x
=
0
{\displaystyle Ax=0}
, luego,
x
∈
Nul
(
A
)
{\displaystyle x\in \operatorname {Nul} (A)}
.