Package 'mvgb'

Title: Multivariate Probabilities of Scale Mixtures of Multivariate Normal Distributions via the Genz and Bretz (2002) QRSVN Method
Description: Generates multivariate subgaussian stable probabilities using the QRSVN algorithm as detailed in Genz and Bretz (2002) <DOI:10.1198/106186002394> but by sampling positive stable variates not chi/sqrt(nu).
Authors: Alan Genz [aut] (wrote mvtdstpack.f), Bruce Swihart [aut, cre] (ORCID: <https://orcid.org/0000-0002-4216-9942>)
Maintainer: Bruce Swihart <[email protected]>
License: LGPL (>= 2.1)
Version: 0.0.6
Built: 2026-06-01 12:00:15 UTC
Source: https://github.com/swihart/mvgb

Help Index

Multivariate Probabilities of Scale Mixtures of Multivariate Normal Distributions

Description

The QRVSN algorithm is used in a broader context: using the Genz and Bretz (2002) algorithm to calculate multivariate distribution probabilities.

Multivariate Subgaussian Stable Distribution

pmvss – multivariate subgaussian stable distribution probabilities

Author(s)

Maintainer: Bruce Swihart [email protected] (ORCID)

Authors:

  • Alan Genz (wrote mvtdstpack.f)

See Also

Useful links:

Multivariate Subgaussian Stable Distribution Probabilities (via positive stable variates)

Description

Computes the the distribution function of the multivariate subgaussian stable distribution for arbitrary limits, alpha, shape matrices, and location vectors. This function is unlike mvtnorm::pmvt in two ways:

  1. The QRVSN method is used for every dimension n (including bivariate and trivariate),

  2. The QRVSN method on positive stable variates, not chi/sqrt(nu).

Usage

pmvss(
  lower,
  upper,
  alpha,
  Q,
  delta = rep(0, NROW(Q)),
  maxpts = 25000,
  abseps = 0.001,
  releps = 0
)

Arguments

lower

lower bounds of integration must be length n, finite
upper

upper bounds of integration must be length n, finite
alpha

real number between 0 and 2 that cannot have more than 6 digits precision. 0.123456 is okay; 0.1234567 is not. Thus alpha in [0.000001, 1.999999].
Q

shape matrix
delta

location vector must have length equal to the number of rows of Q. Defaults to the 0 vector.
maxpts

(description from FORTRAN code) INTEGER, maximum number of function values allowed. This parameter can be used to limit the time. A sensible strategy is to start with MAXPTS = 1000*N, and then increase MAXPTS if ERROR is too large. (description from mvtnorm::GenzBretz) maximum number of function values as integer. The internal FORTRAN code always uses a minimum number depending on the dimension. (for example 752 for three-dimensional problems).
abseps

absolute error tolerance
releps

relative error tolerance as double.

Value

Returns the variables from the MVTDST function (QRVSN algorithm): 

N      INTEGER, the number of variables.

NU     INTEGER, the number of degrees of freedom.
       If NU < 1, then an MVN probability is computed.

LOWER  DOUBLE PRECISION, array of lower integration limits.

UPPER  DOUBLE PRECISION, array of upper integration limits.

INFIN  INTEGER, array of integration limits flags:
        if INFIN(I) < 0, Ith limits are (-infinity, infinity);
        if INFIN(I) = 0, Ith limits are (-infinity, UPPER(I)];
        if INFIN(I) = 1, Ith limits are [LOWER(I), infinity);
        if INFIN(I) = 2, Ith limits are [LOWER(I), UPPER(I)].

CORREL DOUBLE PRECISION, array of correlation coefficients;
       the correlation coefficient in row I column J of the
       correlation matrix should be stored in
          CORREL( J + ((I-2)*(I-1))/2 ), for J < I.
       The correlation matrix must be positive semi-definite.

DELTA  DOUBLE PRECISION, array of non-centrality parameters.

MAXPTS INTEGER, maximum number of function values allowed. This
       parameter can be used to limit the time. A sensible
       strategy is to start with MAXPTS = 1000*N, and then
       increase MAXPTS if ERROR is too large.

ABSEPS DOUBLE PRECISION absolute error tolerance.

RELEPS DOUBLE PRECISION relative error tolerance.

error  DOUBLE PRECISION estimated absolute error,
       with 99% confidence level.

value  DOUBLE PRECISION estimated value for the integral

inform INTEGER, termination status parameter:
       if INFORM = 0, normal completion with ERROR < EPS;
       if INFORM = 1, completion with ERROR > EPS and MAXPTS
                      function values used; increase MAXPTS to
                      decrease ERROR;
       if INFORM = 2, N > 1000 or N < 1.
       if INFORM = 3, correlation matrix not positive semi-definite.

RND    ignore; this initializes RNG

References

Genz, A. and Bretz, F. (2002), Methods for the computation of multivariate t-probabilities. Journal of Computational and Graphical Statistics, 11, 950–971.

http://www.math.wsu.edu/faculty/genz/homepage

http://www.math.wsu.edu/faculty/genz/software/fort77/mvtdstpack.f

Examples

Q = structure(c(1, 0.85, 0.85, 0.85, 0.85,
               0.85, 1   , 0.85, 0.85, 0.85,
               0.85, 0.85   , 1   , 0.85, 0.85,
               0.85, 0.85   , 0.85   , 1   , 0.85,
               0.85, 0.85   , 0.85   , 0.85   , 1   ),
             .Dim = c(5L,5L))

## default maxpts=25000 doesn't finish with error < abseps
mvgb::pmvss(lower=rep(-1,5),
           upper=rep(1,5),
           alpha=1,
           Q=Q,
           maxpts=25000)[c("value","inform","error")]

## increase maxpts to get inform value 0 (that is, error < abseps)
mvgb::pmvss(lower=rep(-1,5),
           upper=rep(1,5),
           alpha=1,
           Q=Q,
           maxpts=25000*350)[c("value","inform","error")]


set.seed(10)
shape_matrix <- structure(c(1, 0.9, 0.9, 0.9, 0.9, 0.9, 1, 0.9, 0.9, 0.9, 0.9,
                           0.9, 1, 0.9, 0.9, 0.9, 0.9, 0.9, 1, 0.9, 0.9, 0.9, 0.9, 0.9,
                           1), .Dim = c(5L, 5L))

mvgb::pmvss(lower=rep(-2,5),
           upper=rep(2,5),
           alpha=1.7,
           Q=shape_matrix,
           delta=rep(0,5),
           maxpts=25000*30)[c("value","inform","error")]