s17def returns a sequence of values for the Bessel functions ${J}_{\nu +n}\left(z\right)$ for complex $z$, non-negative $\nu $ and $n=0,1,\dots ,N-1$, with an option for exponential scaling.
The routine may be called by the names s17def or nagf_specfun_bessel_j_complex.
3Description
s17def evaluates a sequence of values for the Bessel function ${J}_{\nu}\left(z\right)$, where $z$ is complex, $-\pi <\mathrm{arg}z\le \pi $, and $\nu $ is the real, non-negative order. The $N$-member sequence is generated for orders $\nu $, $\nu +1,\dots ,\nu +N-1$. Optionally, the sequence is scaled by the factor ${e}^{-\left|\mathrm{Im}\left(z\right)\right|}$.
Note: although the routine may not be called with $\nu $ less than zero, for negative orders the formula ${J}_{-\nu}\left(z\right)={J}_{\nu}\left(z\right)\mathrm{cos}\left(\pi \nu \right)-{Y}_{\nu}\left(z\right)\mathrm{sin}\left(\pi \nu \right)$ may be used (for the Bessel function ${Y}_{\nu}\left(z\right)$, see s17dcf).
The routine is derived from the routine CBESJ in Amos (1986). It is based on the relations ${J}_{\nu}\left(z\right)={e}^{\nu \pi i/2}{I}_{\nu}(-iz)$, $\mathrm{Im}\left(z\right)\ge 0.0$, and ${J}_{\nu}\left(z\right)={e}^{-\nu \pi i/2}{I}_{\nu}\left(iz\right)$, $\mathrm{Im}\left(z\right)<0.0$.
The Bessel function ${I}_{\nu}\left(z\right)$ is computed using a variety of techniques depending on the region under consideration.
When $N$ is greater than $1$, extra values of ${J}_{\nu}\left(z\right)$ are computed using recurrence relations.
For very large $\left|z\right|$ or $(\nu +N-1)$, argument reduction will cause total loss of accuracy, and so no computation is performed. For slightly smaller $\left|z\right|$ or $(\nu +N-1)$, the computation is performed but results are accurate to less than half of machine precision. If $\mathrm{Im}\left(z\right)$ is large, there is a risk of overflow and so no computation is performed. In all the above cases, a warning is given by the routine.
Amos D E (1986) Algorithm 644: A portable package for Bessel functions of a complex argument and non-negative order ACM Trans. Math. Software12 265–273
5Arguments
1: $\mathbf{fnu}$ – Real (Kind=nag_wp)Input
On entry: $\nu $, the order of the first member of the sequence of functions.
Constraint:
${\mathbf{fnu}}\ge 0.0$.
2: $\mathbf{z}$ – Complex (Kind=nag_wp)Input
On entry: the argument $z$ of the functions.
3: $\mathbf{n}$ – IntegerInput
On entry: $N$, the number of members required in the sequence ${J}_{\nu}\left(z\right),{J}_{\nu +1}\left(z\right),\dots ,{J}_{\nu +N-1}\left(z\right)$.
Constraint:
${\mathbf{n}}\ge 1$.
4: $\mathbf{scal}$ – Character(1)Input
On entry: the scaling option.
${\mathbf{scal}}=\text{'U'}$
The results are returned unscaled.
${\mathbf{scal}}=\text{'S'}$
The results are returned scaled by the factor ${e}^{-\left|\mathrm{Im}\left(z\right)\right|}$.
Constraint:
${\mathbf{scal}}=\text{'U'}$ or $\text{'S'}$.
On exit: the $N$ required function values: ${\mathbf{cy}}\left(i\right)$ contains
${J}_{\nu +i-1}\left(z\right)$, for $\mathit{i}=1,2,\dots ,N$.
6: $\mathbf{nz}$ – IntegerOutput
On exit: the number of components of cy that are set to zero due to underflow. If ${\mathbf{nz}}>0$, then elements ${\mathbf{cy}}\left({\mathbf{n}}-{\mathbf{nz}}+1\right)$, ${\mathbf{cy}}\left({\mathbf{n}}-{\mathbf{nz}}+2\right),\dots ,{\mathbf{cy}}\left({\mathbf{n}}\right)$ are set to zero.
7: $\mathbf{ifail}$ – IntegerInput/Output
On entry: ifail must be set to $0$, $\mathrm{-1}$ or $1$ to set behaviour on detection of an error; these values have no effect when no error is detected.
A value of $0$ causes the printing of an error message and program execution will be halted; otherwise program execution continues. A value of $\mathrm{-1}$ means that an error message is printed while a value of $1$ means that it is not.
If halting is not appropriate, the value $\mathrm{-1}$ or $1$ is recommended. If message printing is undesirable, then the value $1$ is recommended. Otherwise, the value $0$ is recommended. When the value $-\mathbf{1}$ or $\mathbf{1}$ is used it is essential to test the value of ifail on exit.
On exit: ${\mathbf{ifail}}={\mathbf{0}}$ unless the routine detects an error or a warning has been flagged (see Section 6).
6Error Indicators and Warnings
If on entry ${\mathbf{ifail}}=0$ or $\mathrm{-1}$, explanatory error messages are output on the current error message unit (as defined by x04aaf).
Errors or warnings detected by the routine:
${\mathbf{ifail}}=1$
On entry, ${\mathbf{fnu}}=\u27e8\mathit{\text{value}}\u27e9$.
Constraint: ${\mathbf{fnu}}\ge 0.0$.
On entry, ${\mathbf{n}}=\u27e8\mathit{\text{value}}\u27e9$.
Constraint: ${\mathbf{n}}\ge 1$.
On entry, scal has an illegal value: ${\mathbf{scal}}=\u27e8\mathit{\text{value}}\u27e9$.
${\mathbf{ifail}}=2$
No computation because $\mathrm{Im}\left({\mathbf{z}}\right)=\u27e8\mathit{\text{value}}\u27e9>\u27e8\mathit{\text{value}}\u27e9$, ${\mathbf{scal}}=\text{'U'}$.
${\mathbf{ifail}}=3$
Results lack precision because $\left|{\mathbf{z}}\right|=\u27e8\mathit{\text{value}}\u27e9>\u27e8\mathit{\text{value}}\u27e9$.
Results lack precision because ${\mathbf{fnu}}+{\mathbf{n}}-1=\u27e8\mathit{\text{value}}\u27e9>\u27e8\mathit{\text{value}}\u27e9$.
${\mathbf{ifail}}=4$
No computation because $\left|{\mathbf{z}}\right|=\u27e8\mathit{\text{value}}\u27e9>\u27e8\mathit{\text{value}}\u27e9$.
No computation because ${\mathbf{fnu}}+{\mathbf{n}}-1=\u27e8\mathit{\text{value}}\u27e9>\u27e8\mathit{\text{value}}\u27e9$.
${\mathbf{ifail}}=5$
No computation – algorithm termination condition not met.
${\mathbf{ifail}}=-99$
An unexpected error has been triggered by this routine. Please
contact NAG.
See Section 7 in the Introduction to the NAG Library FL Interface for further information.
${\mathbf{ifail}}=-399$
Your licence key may have expired or may not have been installed correctly.
See Section 8 in the Introduction to the NAG Library FL Interface for further information.
${\mathbf{ifail}}=-999$
Dynamic memory allocation failed.
See Section 9 in the Introduction to the NAG Library FL Interface for further information.
7Accuracy
All constants in s17def are given to approximately $18$ digits of precision. Calling the number of digits of precision in the floating-point arithmetic being used $t$, then clearly the maximum number of correct digits in the results obtained is limited by $p=\mathrm{min}\phantom{\rule{0.125em}{0ex}}(t,18)$. Because of errors in argument reduction when computing elementary functions inside s17def, the actual number of correct digits is limited, in general, by $p-s$, where $s\approx \mathrm{max}\phantom{\rule{0.125em}{0ex}}(1,\left|{\mathrm{log}}_{10}\left|z\right|\right|,\left|{\mathrm{log}}_{10}\nu \right|)$ represents the number of digits lost due to the argument reduction. Thus the larger the values of $\left|z\right|$ and $\nu $, the less the precision in the result. If s17def is called with ${\mathbf{n}}>1$, then computation of function values via recurrence may lead to some further small loss of accuracy.
If function values which should nominally be identical are computed by calls to s17def with different base values of $\nu $ and different ${\mathbf{n}}$, the computed values may not agree exactly. Empirical tests with modest values of $\nu $ and $z$ have shown that the discrepancy is limited to the least significant $3$ – $4$ digits of precision.
8Parallelism and Performance
Background information to multithreading can be found in the Multithreading documentation.
s17def is not threaded in any implementation.
9Further Comments
The time taken for a call of s17def is approximately proportional to the value of ${\mathbf{n}}$, plus a constant. In general it is much cheaper to call s17def with ${\mathbf{n}}$ greater than $1$, rather than to make $N$ separate calls to s17def.
Paradoxically, for some values of $z$ and $\nu $, it is cheaper to call s17def with a larger value of ${\mathbf{n}}$ than is required, and then discard the extra function values returned. However, it is not possible to state the precise circumstances in which this is likely to occur. It is due to the fact that the base value used to start recurrence may be calculated in different regions for different ${\mathbf{n}}$, and the costs in each region may differ greatly.
Note that if the function required is ${J}_{0}\left(x\right)$ or ${J}_{1}\left(x\right)$, i.e., $\nu =0.0$ or $1.0$, where $x$ is real and positive, and only a single unscaled function value is required, then it may be much cheaper to call s17aefors17aff respectively.
10Example
This example prints a caption and then proceeds to read sets of data from the input data stream. The first datum is a value for the order fnu, the second is a complex value for the argument, z, and the third is a character value
to set the argument scal. The program calls the routine with ${\mathbf{n}}=2$ to evaluate the function for orders fnu and ${\mathbf{fnu}}+1$, and it prints the results. The process is repeated until the end of the input data stream is encountered.