(i)⇒(ii). Let ε>0; according to (i), there exists m1>0 such that xa+εf(x)<1,∀x>m1 and m2>0 such that xa−εf(x)>1,∀x>m2. If we denote m=max{m1,m2,1}, then xa−ε<f(x)<xa+ε,∀x>m. This implies a−ε<lnxlnf(x)<a+ε,∀x>m. Since ε>0 is arbitrarily chosen, it follows that limx→∞lnxlnf(x)=a.
(ii)⇒(i). Let ε>0. From (ii), we derive the existence of some m>1 such that
a−2ε<lnxlnf(x)<a+2ε,∀x>m.
This yields xa−ε/2<f(x)<xa+ε/2,∀x>m, and therefore xa+εf(x)<xε/21, for all x>m, and xa−εf(x)>xε/2,∀x>m. But limx→∞xε/2=∞ and f(x)>0,∀x>0, from which we obtain the desired conclusion.