Let $A:\mathbb{R}^m \to \mathbb{R}^n$ be a linear transformation. We know that there is a unique transformation $A^*:\mathbb{R}^n \to \mathbb{R}^m$ such that
$$\langle Ax,y\rangle = \langle x,A^*y \rangle, \forall x \in \mathbb{R}^m,y\in \mathbb{R}^n.$$
And we know also that the matrix of $A^*$ is $A^t$.
Having this, prove that the equation
$$Ax=b$$
has a solution if, and only if, $b$ is orthogonal to all of the vectors $y \in \operatorname{ker}A^*.$ Note that this is precisely
$$\operatorname{Im} A = (\operatorname{ker} A^*)^\perp.$$
To show that $\operatorname{Im}A \subset (\operatorname{ker}A^*)^\perp$ is easy.
If $b \in \operatorname{Im}A$ then $\exists x\in \mathbb{R}^m$ such that $Ax=b$. Then, if $y$ is any vector in $\operatorname{ker}A^*$, we have
$$\langle b,y\rangle = \langle Ax,y\rangle$$
$$=\langle x,A^*y\rangle$$
$$=\langle x,0\rangle=0$$
and, therefore, $b \in (\operatorname{ker}A^*)^\perp$, i.e.
$$\operatorname{Im}A \subset (\operatorname{ker}A^*)^\perp.$$
Now, how to show that, conversely, if $b \in (\operatorname{ker}A^*)^\perp$ then $b \in \operatorname{Im}A$??
Need some help... Sorry for duplicates... (I couldn't find this question over here...)

- 2,685
-
2You must have that $\text{Im} \ A$ is a closed space. You can easily verify that $(\text{Im} A)^\perp \subset \text{ker} A^$, hence by taking orthogonal complements you get $( \text{ker} \ A^)^\perp \subset( (\text{Im} \ A)^\perp)^\perp = \overline{\text{Im}\ A}$. Hence only in the case that $\text{Im} \ A$ is a closed space your claim follows, that is the equation $Ax=b$ has a solution when $\text{Im} \ A$ is closed. – Alonso Delfín Sep 03 '15 at 01:49
-
@AlonsoDelfín And $\text{Im} , A$ will be closed, because everything here is finite dimensional. – Stephen Montgomery-Smith Sep 03 '15 at 03:01
-
To paraphrase the argument of @AlonsoDelfín, it is quite easy to prove $(\text{Im},A)^\perp = \text{ker} , A^*$. But to prove what you are looking for, you have to invoke $S^{\perp \perp} = S$ for any subspace $S$. (And if you are working in infinite dimensions, you need that $S$ is closed, but that doesn't apply in your case.) – Stephen Montgomery-Smith Sep 03 '15 at 03:05
-
@StephenMontgomery-Smith you are right, I did not look that we were working on finite dimensional spaces, thus indeed $\text{Im} \ A$ is a closed set. – Alonso Delfín Sep 03 '15 at 04:44
1 Answers
Let $b$ be any element in $(\operatorname{ker}A^*)^\perp$.
Since $\mathbb{R}^n$ has finite dimension
$$\mathbb{R}^n=\operatorname{Im}A \oplus (\operatorname{Im}A)^\perp$$
holds, and we can write $b=Au+w$, $u\in \mathbb{R}^m$, $w \in (\operatorname{Im}A)^\perp$. We shall be done if we show that $w=0$ and, therefore, $b\in \operatorname{Im}A$.
Now, for any $v \in \mathbb{R}^m$, we have
$$\langle A^*w,v\rangle=\langle w,Av\rangle=0,$$
since $w\in (\operatorname{Im}A)^\perp$. This shows that $A^*w=0$ (take $v=A^*w$). So, $w \in \operatorname{ker}A^*$.
Finally, how $b\in(\operatorname{ker}A^*)^\perp$, we have
$$\langle b,w\rangle=0.$$
On the other hand,
$$\begin{array}{rcl}
\langle b,w\rangle&=&\langle Au+w,w \rangle\\
&=& \langle Au,w\rangle+\langle w,w \rangle\\
&=&\langle w,w\rangle\end{array}$$
And hence $\langle w,w \rangle =0$ and we are done.

- 2,685