Using Markov's interval arithmetic to evaluate Bessel-Riccati functions (Q1904147)

The authors consider Bessel-Riccati functions which are given by \[ \psi_n(x)= \sqrt x J_{n+{1\over 2}},\quad \chi_n(x)= \sqrt x Y_{n+ {1\over 2}}, \] where \(J_{n+ {1\over 2}}\) and \(Y_{n+ {1\over 2}}\) are half-order Bessel functions of the first and of the second kind, respectively. The subject of the paper consists in enclosing the range of these functions over compact intervals by applying interval arithmetic tools. To this end, five methods are used: The first is based on the standard interval arithmetic evaluation, the second uses the mean value form, the third involves Taylor expansion up to the polynomial degree one with bounds for the second-order remainder term, the fourth pursues a bisection strategy for the starting interval, and the fifth applies an extended interval arithmetic due to S. Markov. In Section 1, interval arithmetic and its extension are defined, and elementary properties are presented. In Section 2 some remarks are given concerning the implementation of the extended interval arithmetic in FORTRAN 90. In addition, the methods and the functions \(\psi_n\) and \(\chi_n\) are introduced. Numerical results are listed and compared. Code in FORTRAN 90 is added. Since for the fifth method monotonicity is exploited Section 3 is devoted to automatic differentiation which is involved in order to prove this property. In Section 4 conclusions are drawn from the numerical results.