Fredholm alternative for a matrix (source code)

= Fredholm alternative for a matrix
{c}

For a complex $m$-by-$n$ <matrix> $A$, the <linear system> $Ax=b$ has a solution precisely when $b$ is <orthogonal> to every <vector> in $\ker A^\dagger$, where $A^\dagger$ is the <conjugate transpose>. For a real <matrix>, use the <transpose> $A^T$ instead. Indeed,
$$
(\operatorname{im}A)^\perp=\ker A^\dagger,
$$
because the defining <adjoint operator> identity turns orthogonality to every $Ax$ into $A^\dagger y=0$. Taking the <orthogonal complement> again recovers $\operatorname{im}A$ in finite dimensions. Thus a <linear functional> in the <annihilator of a vector subspace> gives the exact obstruction to solvability, rather than just a necessary test.

For $A=\operatorname{diag}(1,0)$, its <image of a linear map> is the first coordinate axis and $\ker A^\dagger$ is the second. Consequently $Ax=(b_1,b_2)^T$ is solvable exactly when $b_2=0$.