Argument, x >= 0 (NaN for negative x).
Order, any real number. For nu < 0, J(x, -nu) = cos(pi nu) J(x, nu) - sin(pi nu) Y(x, nu)
(A&S 9.1.2, as in R).
J_nu(x).
For x much larger than nu but below the range of the asymptotic expansion (x < nu / 0.011), the time grows
with x (a continued fraction of about x - nu terms): about 15 ms at x = 5e4, nu = 1000, 1.6 s at x = 1e7,
nu = 2e5. Near the turning point (x a little below nu) a recurrence takes about nu steps. Beyond 2^25 terms
of the continued fraction or 2^31 steps of the recurrence (2^28 for a negative non-integer order, which needs Y) a
RangeError is thrown before the computation. So it is in the asymptotic expansion (x above about 90 nu) where
the phase correction nu^2 / (2x) exceeds 2^26, which double precision does not resolve.
Bessel function of the first kind. Equivalent to R's
besselJ(x, nu). Takes single numbers.