2
0
mirror of https://github.com/boostorg/math.git synced 2026-01-22 05:22:15 +00:00
Files
math/doc/beta.qbk
2006-06-20 18:04:51 +00:00

109 lines
3.6 KiB
Plaintext

[#beta][section The Beta Function]
[caution __caution ]
[h4 Synopsis]
``
#include <boost/math/special_functions/beta.hpp>
``
namespace boost{ namespace math{
template <class T>
T beta(T a, T b);
}} // namespaces
[h4 Description]
The beta function is defined by:
[$../equations/beta1.png]
[$../graphs/beta.png]
And for small values:
[$../graphs/beta-small.png]
There are effectively two versions of this function internally: a fully
generic version that is slow, but reasonably accurate, and a much more
efficient approximation that is used where the number of digits in the mantissa
of T correspond to a certain __lanczos. In practice any built in
floating point type you will encounter has an appropriate __lanczos
defined for it. It is also possible, given enough machine time, to generate
further __lanczos's using the program libs/math/tools/lanczos_generator.cpp.
[h4 Accuracy]
The following table shows peak errors for various domains of input arguments,
plus comparisons to other open source implementations where available. Note that
only results for the widest floating point type on the system are given as
narrower types have __zero_error.
[table Peak Errors In the Beta Function
[[Mantissa Size] [Platform and Compiler] [Errors in range\n0.4 < a,b < 100] [Errors in range\n1e-6 < a,b < 36]]
[[53] [Win32, Visual C++ 8] [Peak=99 Mean=22\n\n(GSL Peak=1178 Mean=238)\n(Cephes=1612)] [Peak=10.7 Mean=2.6\n\n(GSL Peak=12 Mean=2.0)\n(Cephes=174)]]
[[64] [Red Hat Linux IA32, g++ 3.4.4] [Peak=112.1 Mean=26.9] [Peak=15.8 Mean=3.6]]
[[64] [Red Hat Linux IA64, g++ 3.4.4] [Peak=61.4 Mean=19.5] [Peak=12.2 Mean=3.6]]
[[113] [HPUX IA64, aCC A.06.06] [Peak=42.03 Mean=13.94] [Peak=9.8 Mean=3.1]]
]
Note that the worst errors occur when a or b are large, and that
when this is the case the result is very close to zero, so absolute
errors will be very small.
[h4 Testing]
A mixture of spot tests of exact values, and randomly generated test data are
used: the test data was computed using NTL::RR at 1000-bit precision.
[h4 Implementation]
Traditional methods of evaluating the beta function either involve evaluating
the gamma functions directly, or taking logarithms and then
exponentiating the result. However, the former is prone to overflows
for even very modest arguments, while the latter is prone to cancellation
errors. As an alternative, if we regard the gamma function as a white-box
containing the __lanczos then we can combine the power terms:
[$../equations/beta2.png]
which is almost the ideal solution, however almost all of the error occurs
in evaluating the power terms when /a/ or /b/ are large. If we assume that /a > b/
then the larger of the two power terms can be reduced by a factor of /b/, which
immediately cuts the maximum error in half:
[$../equations/beta3.png]
This may not be the final solution, but it is very competitive compared to
other implementation methods.
The generic implementation - where no __lanczos approximation is available - is
implemented in a very similar way to the generic version of the gamma function.
Again in order to avoid numerical overflow the power terms that prefix the series and
continued fraction parts are collected together into:
[$../equations/beta8.png]
where la, lb and lc are the integration limits used for a, b, and a+b.
There are a few special cases worth mentioning:
When /a/ or /b/ are less than one, we can use the recurrence relations:
[$../equations/beta4.png]
[$../equations/beta5.png]
To move to a more favorable region where they are both greater than 1.
In addition:
[$../equations/beta7.png]
[endsect]