theorem phieff_exceeds_phi0 {phi_0 beta_auto : ℝ}
(hphi : 0 < phi_0) (hbeta : 0 < beta_auto)
(hprod : beta_auto * phi_0 < 1) :
phi_0 < effectiveTrainingProductivity phi_0 beta_auto := by
unfold effectiveTrainingProductivity
have hdenom_pos : 0 < 1 - beta_auto * phi_0 := by linarith
rw [lt_div_iff₀ hdenom_pos]
have : 0 < beta_auto * phi_0 := mul_pos hbeta hphi
nlinarithPaper 6: Endogenous Decentralization and the AI Transition