\@doendnote{endnote43}{Due to Furry's theorem the two-loop singlet contribution is zero for the vector current.} \@doendnote{endnote44}{Note that in Ref.\protect \nobreakspace {}\cite {Marquard:2006qi} the result has been expressed in terms of the coupling defined in the full theory whereas here we use the effective one denoted by $\alpha _s^{(n_l)}$. This explains the difference in the logarithmic part of the coefficient $c_{FHL}$.} \@doendnote{endnote45}{Note that in Ref.\protect \nobreakspace {}\cite {Marquard:2009bj}, where the fermionic contributions are given, only a factor two has been chosen which explains the slight increase of the uncertainty of $c_{FAH}$ in Eq.\protect \nobreakspace {}(\ref {eq::cv3}).} \@doendnote{endnote46}{Note that there is a misprint in Eq.\protect \nobreakspace {}(5) of Ref.\protect \nobreakspace {}\cite {Beneke:2007uf}: the term $E(1-d_v/3)/m$ should read $E(c_v-d_v/3)/m$.}