[section:gamma_dist Gamma (and Erlang) Distribution]

``#include <boost/math/distributions/gamma.hpp>``

   namespace boost{ namespace math{ 
      
   template <class RealType = double, 
             class ``__Policy``   = ``__policy_class`` >
   class gamma_distribution
   {
   public:
      typedef RealType value_type;
      typedef Policy   policy_type;

      BOOST_MATH_GPU_ENABLED gamma_distribution(RealType shape, RealType scale = 1)

      BOOST_MATH_GPU_ENABLED RealType shape()const;
      BOOST_MATH_GPU_ENABLED RealType scale()const;
   };
   
   }} // namespaces
   
The gamma distribution is a continuous probability distribution.
When the shape parameter is an integer then it is known as the 
Erlang Distribution.  It is also closely related to the Poisson
and Chi Squared Distributions.

When the shape parameter has an integer value, the distribution is the
[@http://en.wikipedia.org/wiki/Erlang_distribution Erlang distribution].
Since this can be produced by ensuring that the shape parameter has an
integer value > 0, the Erlang distribution is not separately implemented.

[note
To avoid potential confusion with the gamma functions, this
distribution does not provide the typedef:

``typedef gamma_distribution<double> gamma;`` 

Instead if you want a double precision gamma distribution you can write 

``boost::math::gamma_distribution<> my_gamma(1, 1);``
]

For shape parameter /k/ and scale parameter [theta] it is defined by the
probability density function:

[equation gamma_dist_ref1]

Sometimes an alternative formulation is used: given parameters
[alpha] = k and [beta] = 1 / [theta], then the 
distribution can be defined by the PDF:

[equation gamma_dist_ref2]

In this form the inverse scale parameter is called a /rate parameter/.

Both forms are in common usage: this library uses the first definition
throughout.  Therefore to construct a Gamma Distribution from a ['rate
parameter], you should pass the reciprocal of the rate as the scale parameter.

The following two graphs illustrate how the PDF of the gamma distribution
varies as the parameters vary:

[graph gamma1_pdf]

[graph gamma2_pdf]

The [*Erlang Distribution] is the same as the Gamma, but with the shape parameter
an integer.  It is often expressed using a /rate/ rather than a /scale/ as the 
second parameter (remember that the rate is the reciprocal of the scale).

Internally the functions used to implement the Gamma Distribution are
already optimised for small-integer arguments, so in general there should
be no great loss of performance from using a Gamma Distribution rather than
a dedicated Erlang Distribution.

[h4 Member Functions]

   BOOST_MATH_GPU_ENABLED gamma_distribution(RealType shape, RealType scale = 1);
   
Constructs a gamma distribution with shape /shape/ and 
scale /scale/.

Requires that the shape and scale parameters are greater than zero, otherwise calls
__domain_error.

   BOOST_MATH_GPU_ENABLED RealType shape()const;
   
Returns the /shape/ parameter of this distribution.
   
   BOOST_MATH_GPU_ENABLED RealType scale()const;
      
Returns the /scale/ parameter of this distribution.

[h4 Non-member Accessors]

All the [link math_toolkit.dist_ref.nmp usual non-member accessor functions] that are generic to all
distributions are supported: __usual_accessors.
For this distribution all non-member accessor functions are marked with `BOOST_MATH_GPU_ENABLED` and can
be run on both host and device.

The domain of the random variable is \[0,+[infin]\].

In this distribution the implementation of `logpdf` is specialized
to improve numerical accuracy.

[h4 Accuracy]

The gamma distribution is implemented in terms of the 
incomplete gamma functions __gamma_p and __gamma_q and their
inverses __gamma_p_inv and __gamma_q_inv: refer to the accuracy
data for those functions for more information.

[h4 Implementation]

In the following table /k/ is the shape parameter of the distribution, 
[theta] is its scale parameter, /x/ is the random variate, /p/ is the probability
and /q = 1-p/.

[table
[[Function][Implementation Notes]]
[[pdf][Using the relation: pdf = __gamma_p_derivative(k, x / [theta]) / [theta] ]]
[[logpdf][log(pdf) = -k*log([theta]) + (k-1)*log(x) - lgamma(k) - (x/[theta]) ]]
[[cdf][Using the relation: p = __gamma_p(k, x / [theta]) ]]
[[cdf complement][Using the relation: q = __gamma_q(k, x / [theta]) ]]
[[quantile][Using the relation: x = [theta] * __gamma_p_inv(k, p) ]]
[[quantile from the complement][Using the relation: x = [theta]* __gamma_q_inv(k, p) ]]
[[mean][k[theta] ]]
[[variance][k[theta][super 2] ]]
[[mode][(k-1)[theta] for ['k>1] otherwise a __domain_error ]]
[[skewness][2 / sqrt(k) ]]
[[kurtosis][3 + 6 / k]]
[[kurtosis excess][6 / k ]]
]

[endsect] [/section:gamma_dist Gamma (and Erlang) Distribution]


[/ 
  Copyright 2006, 2010 John Maddock and Paul A. Bristow.
  Distributed under the Boost Software License, Version 1.0.
  (See accompanying file LICENSE_1_0.txt or copy at
  http://www.boost.org/LICENSE_1_0.txt).
]