If I understand correctly, ftol and xtol currently set the absolute error tolerance for the residual vectors. Is there a way to use a relative error instead, in case the target function and/or its input vector are relatively large and convergence up to several digits might be difficult to achieve?
If I understand correctly,
ftol
andxtol
currently set the absolute error tolerance for the residual vectors. Is there a way to use a relative error instead, in case the target function and/or its input vector are relatively large and convergence up to several digits might be difficult to achieve?