The three-term recursion for Chebyshev polynomials is mixed forward-backward stable (Q494669)

An algorithm \(W\) for computing the Chebyshev polynomial \(T_N(x)\) is called mixed forward-backward stable with respect to the data \(x\) if the value \(\widetilde T_N(x)\) computed by \(W\) in floating point arithmetic satisfies \[ \widetilde T_N(x)= (1 + \delta_N)T_N((1 + \Delta_N)x) + O(\epsilon_M^2), \quad |\delta_N|, |\Delta_N| \leq \epsilon_ML, \] where \(L=L(N)\) is a modestly growing function on \(N\) and \(\epsilon_M\) is the machine precision. The authors prove that the three-term recurrence relation \(T_{n+1}(x)=2xT_n(x) - T_{n-1}(x)\) is mixed forward-backward stable.