math/doc/distributions/inverse_chi_squared.qbk

[section:inverse_chi_squared_dist Inverse Chi Squared Distribution]

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

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

      inverse_chi_squared_distribution(RealType df = 1); // Not explicitly scaled, default 1/df.
      inverse_chi_squared_distribution(RealType df, RealType scale = 1/df);  // Scaled.

      RealType degrees_of_freedom()const; // Default 1.
      RealType scale()const; // Optional scale [xi] (variance), default 1/degrees_of_freedom.
   };
   
   }} // namespace boost // namespace math
   
The inverse chi squared distribution is a continuous probability distribution
of the *reciprocal* of a variable distributed according to the chi squared distribution.

The sources below give confusingly different formulae
using different symbols for the distribution pdf,
but they are all the same, or related by a change of variable, or choice of scale.

Two constructors are available to implement both the scaled and (implicitly) unscaled versions.

The main version has an explicit scale parameter which implements the
[@http://en.wikipedia.org/wiki/Scaled-inverse-chi-square_distribution scaled inverse chi_squared distribution].

A second version has an implicit scale = 1/degrees of freedom and gives the 1st definition in the
[@http://en.wikipedia.org/wiki/Inverse-chi-square_distribution Wikipedia inverse chi_squared distribution].
The 2nd Wikipedia inverse chi_squared distribution definition can be implemented
by  explicitly specifying a scale = 1.

Both definitions are also available in __Mathematica and in __R (geoR) with default scale = 1/degrees of freedom.

See 

* Inverse chi_squared distribution [@http://en.wikipedia.org/wiki/Inverse-chi-square_distribution]
* Scaled inverse chi_squared distribution[@http://en.wikipedia.org/wiki/Scaled-inverse-chi-square_distribution] 
* R inverse chi_squared distribution functions [@http://hosho.ees.hokudai.ac.jp/~kubo/Rdoc/library/geoR/html/InvChisquare.html R ]
* Inverse chi_squared distribution functions [@http://mathworld.wolfram.com/InverseChi-SquaredDistribution.html Weisstein, Eric W. "Inverse Chi-Squared Distribution." From MathWorld--A Wolfram Web Resource.] 
* Inverse chi_squared distribution reference [@http://reference.wolfram.com/mathematica/ref/InverseChiSquareDistribution.html Weisstein, Eric W. "Inverse Chi-Squared Distribution reference." From Wolfram Mathematica.]

The inverse_chi_squared distribution is used in
[@http://en.wikipedia.org/wiki/Bayesian_statistics Bayesian statistics]:
the scaled inverse chi-square is conjugate prior for the normal distribution
with known mean, model parameter [sigma][pow2] (variance).

See [@http://en.wikipedia.org/wiki/Conjugate_prior conjugate priors including a table of distributions and their priors.]

See also __inverse_gamma_distrib and __chi_squared_distrib.

The inverse_chi_squared distribution is a special case of a inverse_gamma distribution
with [nu] (degrees_of_freedom) shape ([alpha]) and scale ([beta]) where

[expression [alpha]= [nu] /2 and [beta] = [frac12]]

[note This distribution *does* provide the typedef:

``typedef inverse_chi_squared_distribution<double> inverse_chi_squared;`` 

If you want a `double` precision inverse_chi_squared distribution you can use 

``boost::math::inverse_chi_squared_distribution<>``

or you can write `inverse_chi_squared my_invchisqr(2, 3);`]

For degrees of freedom parameter [nu],
the (*unscaled*) inverse chi_squared distribution is defined by the probability density function (PDF):

[expression f(x;[nu]) = 2[super -[nu]/2] x[super -[nu]/2-1] e[super -1/2x] / [Gamma]([nu]/2)]

and Cumulative Density Function (CDF)

[expression F(x;[nu]) = [Gamma]([nu]/2, 1/2x) / [Gamma]([nu]/2)]

For degrees of freedom parameter [nu] and scale parameter [xi],
the *scaled* inverse chi_squared distribution is defined by the probability density function (PDF):

[expression f(x;[nu], [xi]) = ([xi][nu]/2)[super [nu]/2] e[super -[nu][xi]/2x] x[super -1-[nu]/2] / [Gamma]([nu]/2)]

and Cumulative Density Function (CDF)

[expression  F(x;[nu], [xi]) = [Gamma]([nu]/2, [nu][xi]/2x) / [Gamma]([nu]/2)]

The following graphs illustrate how the PDF and CDF of the inverse chi_squared distribution
varies for a few values of parameters [nu] and [xi]:

[graph inverse_chi_squared_pdf]  [/.png or .svg]

[graph inverse_chi_squared_cdf]

[h4 Member Functions]

   inverse_chi_squared_distribution(RealType df = 1); // Implicitly scaled 1/df.
   inverse_chi_squared_distribution(RealType df = 1, RealType scale); // Explicitly scaled.

Constructs an inverse chi_squared distribution with [nu] degrees of freedom ['df],
and scale ['scale] with default value 1\/df.

Requires that the degrees of freedom [nu] parameter is greater than zero, otherwise calls
__domain_error.

   RealType degrees_of_freedom()const; 
   
Returns the degrees_of_freedom [nu] parameter of this distribution.

   RealType scale()const; 
   
Returns the scale [xi] 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.

The domain of the random variate is \[0,+[infin]\].
[note Unlike some definitions, this implementation supports a random variate 
equal to zero as a special case, returning zero for both pdf and cdf.]

[h4 Accuracy]

The inverse gamma distribution is implemented in terms of the 
incomplete gamma functions like the __inverse_gamma_distrib that use 
__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.
But in general, gamma (and thus inverse gamma) results are often accurate to a few epsilon,
>14 decimal digits accuracy for 64-bit double.
unless iteration is involved, as for the estimation of degrees of freedom.

[h4 Implementation]

In the following table [nu] is the degrees of freedom parameter and 
[xi] is the scale parameter of the distribution,
/x/ is the random variate, /p/ is the probability and /q = 1-p/ its complement.
Parameters [alpha] for shape and [beta] for scale
are used for the inverse gamma function: [alpha] = [nu]/2 and [beta] = [nu] * [xi]/2.

[table
[[Function][Implementation Notes]]
[[pdf][Using the relation: pdf = __gamma_p_derivative([alpha], [beta]/ x, [beta]) / x * x ]]
[[cdf][Using the relation: p = __gamma_q([alpha], [beta] / x) ]]
[[cdf complement][Using the relation: q = __gamma_p([alpha], [beta] / x) ]]
[[quantile][Using the relation: x = [beta]/ __gamma_q_inv([alpha], p) ]]
[[quantile from the complement][Using the relation: x = [alpha]/ __gamma_p_inv([alpha], q) ]]
[[mode][[nu] * [xi] / ([nu] + 2) ]]
[[median][no closed form analytic equation is known, but is evaluated as quantile(0.5)]]
[[mean][[nu][xi] / ([nu] - 2) for [nu] > 2, else a __domain_error]]
[[variance][2 [nu][pow2] [xi][pow2] / (([nu] -2)[pow2] ([nu] -4)) for [nu] >4, else a __domain_error]]
[[skewness][4 [sqrt]2 [sqrt]([nu]-4) /([nu]-6) for [nu] >6, else a __domain_error ]]
[[kurtosis_excess][12 * (5[nu] - 22) / (([nu] - 6) * ([nu] - 8)) for [nu] >8, else a __domain_error]]
[[kurtosis][3 + 12 * (5[nu] - 22) / (([nu] - 6) * ([nu]-8)) for [nu] >8, else a __domain_error]]
] [/table]

[h4 References]

# Bayesian Data Analysis, Andrew Gelman, John B. Carlin, Hal S. Stern, Donald B. Rubin,
ISBN-13: 978-1584883883, Chapman & Hall; 2 edition (29 July 2003).

# Bayesian Computation with R, Jim Albert, ISBN-13: 978-0387922973, Springer; 2nd ed. edition (10 Jun 2009)

[endsect] [/section:inverse_chi_squared_dist Inverse chi_squared Distribution]

[/ 
  Copyright 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).
]