This is probably a dumb question; so far, I have been using $g^{ij}g_{ij} = n$ as granted, but I am probably doing something wrong here. Suppose that $g_{ij}(t)$ is a time-evolving metric. Let $h_{ij} = \partial_t g_{ij}(t)$. In order to find $\partial_t(g^{ij})$, I need to do:
\begin{alignat*}{2} &\quad \partial_t (g^{ij}g_{jk}) &&= 0 \\ \Leftrightarrow& \quad g_{jk}(\partial_tg^{ij}) + g^{ij}(\partial_t g_{jk}) &&= 0 \\ \Leftrightarrow& \quad g_{jk}(\partial_tg^{ij}) &&= -g^{ij}(\partial_t g_{jk}) = -g^{ij} h_{jk} = -h^i_k \end{alignat*} At this step, standard thing would be to multiply each side by $g^{kl}$ so that I get $\partial_t g^{il} = -h^{il}$. However, what happens if I multiply by $g^{jk}$?
\begin{alignat*}{2} & g^{jk}g_{jk}(\partial_tg^{ij}) && = -g^{jk}h^i_k \\ \Leftrightarrow \quad& n(\partial_t g^{ij}) = -h^{ij} \end{alignat*}
I am assuming the former is clearly right. But I don't see an error with multiplying both sides by $g^{jk}$ instead of $g^{kl}$. What is the error that I am making?