= Frobenius inner product
{c}
{title2=$\langle A,B\rangle_F=\operatorname{tr}(A^*B)$}
= Trace inner product
{synonym}
For complex $m\times n$ <matrices>, the <Frobenius inner product> is
$$
\langle A,B\rangle_F=\operatorname{tr}(A^*B)=\sum_{i=1}^m\sum_{j=1}^n\overline{a_{ij}}b_{ij}.
$$
Here $A^*$ is the <conjugate transpose>, and this convention is conjugate-linear in the first argument. It is the ordinary Euclidean <inner product> on all matrix entries, so $\langle A,A\rangle_F=\sum_{i,j}|a_{ij}|^2$ is positive for every nonzero <matrix>. Its induced <norm> is the <Frobenius norm>.
On real <symmetric matrices> it becomes $\operatorname{tr}(AB)$. In particular, for a real <symmetric matrix> $A$ and a real <vector> $x$,
$$
\langle A,xx^T\rangle_F=\operatorname{tr}(Axx^T)=x^TAx.
$$
Thus a <quadratic form> evaluated at $x$ is a linear test of $A$ against a rank-one <positive semidefinite matrix>. This connects the geometry of <matrices> with quadratic measurements of <vectors>.
Back to article page