In QCD, and more generally in representations of $\mathfrak{su}(N)$, there is a freedom to choose the normalisation of the generators, $$ \mathrm{Tr} \, \left[R(T^a) R(T^b)\right] = T_R \delta^{ab}.\tag{1} $$
I am trying to work out the implications of this for the kinetic term for the gluon field in the QCD Lagrangian density. I've looked in various textbooks and notes, and every source I can find either glosses over this entirely, or gets the details trivially wrong (e.g. eqn. 33 in these or eqn. 4.66 in these).
Conventionally physicists choose $ T_F = \frac{1}{2}$ and write $$ \mathcal{L}_\mathrm{kin} = - \frac{1}{4} F_{\mu\nu}^a F^{\mu\nu}{}^a,\tag{2} $$ where the field strength tensor has been expanded into components $$ \mathbf{F}_{\mu\nu} = \sum_a F_{\mu\nu}^a T^a.\tag{3} $$
It seems clear to me that $$ \mathrm{Tr} \, \left[\mathbf{F}_{\mu\nu} \mathbf{F}^{\mu\nu} \right] = T_F \; F_{\mu\nu}^a F^{\mu\nu}{}^a,\tag{4} $$ and so the correct, Lie-algebra-normalisation-convention-independent expression for the Lagrangian density should be $$ \mathcal{L}_\mathrm{kin} = - \frac{1}{2} \mathrm{Tr} \, \left[\mathbf{F}_{\mu\nu} \mathbf{F}^{\mu\nu} \right] = - \frac{1}{2} T_F \; F_{\mu\nu}^a F^{\mu\nu}{}^a,\tag{5} $$ which restores the usual result in the conventional case.
Is this correct, and is there a good source for it? (ie one that could be cited in an academic work).
Why is the convention-specific component expansion used preferentially in the literature?