quantum-espresso/Modules/random_numbers.f90

288 lines
8.8 KiB
Fortran

!
! Copyright (C) 2001-2012 Quantum ESPRESSO group
! This file is distributed under the terms of the
! GNU General Public License. See the file `License'
! in the root directory of the present distribution,
! or http://www.gnu.org/copyleft/gpl.txt .
!
!----------------------------------------------------------------------------
MODULE random_numbers
!----------------------------------------------------------------------------
!! Module for random numbers generation.
!
USE kinds, ONLY : DP
!
IMPLICIT NONE
!
INTERFACE gauss_dist
!
MODULE PROCEDURE gauss_dist_scal, gauss_dist_vect
!
END INTERFACE
!
CONTAINS
!
!------------------------------------------------------------------------
FUNCTION randy ( irand )
!------------------------------------------------------------------------
!! * x=randy(n): reseed with initial seed \(\text{idum}=n\) ( 0 <= n <= ic, see below)
!! if randy is not explicitly initialized, it will be
!! initialized with seed \(\text{idum}=0\) the first time it is called;
!! * x=randy() : generate uniform REAL(DP) numbers x in [0,1].
!
REAL(DP) :: randy
INTEGER, optional :: irand
!
INTEGER , PARAMETER :: m = 714025, &
ia = 1366, &
ic = 150889, &
ntab = 97
REAL(DP), PARAMETER :: rm = 1.0_DP / m
INTEGER :: j
INTEGER, SAVE :: ir(ntab), iy, idum=0
LOGICAL, SAVE :: first=.true.
!
IF ( present(irand) ) THEN
idum = MIN( ABS(irand), ic)
first=.true.
END IF
IF ( first ) THEN
!
first = .false.
idum = MOD( ic - idum, m )
!
DO j=1,ntab
idum=mod(ia*idum+ic,m)
ir(j)=idum
END DO
idum=mod(ia*idum+ic,m)
iy=idum
END IF
j=1+(ntab*iy)/m
IF( j > ntab .OR. j < 1 ) call errore('randy','j out of range',ABS(j)+1)
iy=ir(j)
randy=iy*rm
idum=mod(ia*idum+ic,m)
ir(j)=idum
!
RETURN
!
END FUNCTION randy
!
!------------------------------------------------------------------------
SUBROUTINE set_random_seed ( )
!------------------------------------------------------------------------
!! poor-man random seed for \(\texttt{randy}\).
!
INTEGER, DIMENSION (8) :: itime
INTEGER :: iseed, irand
!
CALL date_and_time ( values = itime )
! itime contains: year, month, day, time difference (minutes) from UTC,
! hours, minutes, seconds and milliseconds.
! The following rather arbitrary choice, modified as suggested by
! Han Hsu, appears to yield sufficiently randomized initial seeds
!
iseed = ( itime(8) + itime(6) ) * ( itime(7) + itime(5) )
irand = randy ( iseed )
!
END SUBROUTINE set_random_seed
!
!-----------------------------------------------------------------------
FUNCTION gauss_dist_scal( mu, sigma )
!-----------------------------------------------------------------------
!! This function generates a number taken from a normal distribution of
!! mean value \(\text{mu}\) and variance \(\text{sigma}.
!
IMPLICIT NONE
!
REAL(DP), INTENT(IN) :: mu
REAL(DP), INTENT(IN) :: sigma
REAL(DP) :: gauss_dist_scal
!
REAL(DP) :: x1, x2, w
!
!
gaussian_loop: DO
!
x1 = 2.0_DP * randy() - 1.0_DP
x2 = 2.0_DP * randy() - 1.0_DP
!
w = x1 * x1 + x2 * x2
!
IF ( w < 1.0_DP ) EXIT gaussian_loop
!
END DO gaussian_loop
!
w = SQRT( ( - 2.0_DP * LOG( w ) ) / w )
!
gauss_dist_scal = x1 * w * sigma + mu
!
RETURN
!
END FUNCTION gauss_dist_scal
!
!-----------------------------------------------------------------------
FUNCTION gauss_dist_cmplx( mu, sigma )
!-----------------------------------------------------------------------
!! This function generates a number taken from a normal distribution of
!! mean value \(\text{mu}\) and variance \(\text{sigma}\).
!
IMPLICIT NONE
!
REAL(DP), INTENT(IN) :: mu
REAL(DP), INTENT(IN) :: sigma
COMPLEX(DP) :: gauss_dist_cmplx
!
REAL(DP) :: x1, x2, w
!
!
gaussian_loop: DO
!
x1 = 2.0_DP * randy() - 1.0_DP
x2 = 2.0_DP * randy() - 1.0_DP
!
w = x1 * x1 + x2 * x2
!
IF ( w < 1.0_DP ) EXIT gaussian_loop
!
END DO gaussian_loop
!
w = SQRT( ( - 2.0_DP * LOG( w ) ) / w )
!
gauss_dist_cmplx = CMPLX( x1 * w * sigma + mu, x2 * w * sigma + mu, kind=DP)
!
RETURN
!
END FUNCTION gauss_dist_cmplx
!
!-----------------------------------------------------------------------
FUNCTION gauss_dist_vect( mu, sigma, dim )
!-----------------------------------------------------------------------
!! This function generates an array of numbers taken from a normal
!! distribution of mean value \(\text{mu}\) and variance \(\text{sigma}\).
!
IMPLICIT NONE
!
REAL(DP), INTENT(IN) :: mu
REAL(DP), INTENT(IN) :: sigma
INTEGER, INTENT(IN) :: dim
REAL(DP) :: gauss_dist_vect( dim )
!
REAL(DP) :: x1, x2, w
INTEGER :: i
!
!
DO i = 1, dim, 2
!
gaussian_loop: DO
!
x1 = 2.0_DP * randy() - 1.0_DP
x2 = 2.0_DP * randy() - 1.0_DP
!
w = x1 * x1 + x2 * x2
!
IF ( w < 1.0_DP ) EXIT gaussian_loop
!
END DO gaussian_loop
!
w = SQRT( ( - 2.0_DP * LOG( w ) ) / w )
!
gauss_dist_vect(i) = x1 * w * sigma
!
IF ( i >= dim ) EXIT
!
gauss_dist_vect(i+1) = x2 * w * sigma
!
END DO
!
gauss_dist_vect(:) = gauss_dist_vect(:) + mu
!
RETURN
!
END FUNCTION gauss_dist_vect
!
!-----------------------------------------------------------------------
FUNCTION gamma_dist (ialpha)
!-----------------------------------------------------------------------
!! Gamma-distributed random number, implemented as described in
!! Numerical recipes (Press, Teukolsky, Vetterling, Flannery).
!
IMPLICIT NONE
INTEGER, INTENT(IN) :: ialpha
REAL(DP) gamma_dist
INTEGER j
REAL(DP) am,e,s,v1,v2,x,y
REAL(DP), external :: ran1
!
IF ( ialpha < 1 ) CALL errore('gamma_dist', 'bad alpha in gamma_dist', 1)
!
! For small alpha, it is more efficient to calculate Gamma as the waiting time
! to the alpha-th event oin a Poisson random process of unit mean.
! Define alpha as small for 0 < alpha < 6:
IF ( ialpha < 6 ) THEN
!
x = 1.0D0
DO j=1,ialpha
x = x * randy()
ENDDO
x = -LOG(x)
ELSE
DO
v1 = 2.0D0*randy()-1.0D0
v2 = 2.0D0*randy()-1.0D0
!
! need to get this condition met:
IF ( v1**2+v2**2 > 1.0D0) CYCLE
!
y = v2 / v1
am = ialpha - 1
s = sqrt(2.0D0 * am + 1.0D0)
x = s * y + am
!
IF ( x <= 0.) CYCLE
!
e = (1.0D0+y**2)* exp( am * log( x / am ) - s * y)
!
IF (randy() > e) THEN
CYCLE
ELSE
EXIT
ENDIF
ENDDO
ENDIF
!
gamma_dist=x
!
ENDFUNCTION gamma_dist
!
!-----------------------------------------------------------------------
FUNCTION sum_of_gaussians2(inum_gaussians)
!-----------------------------------------------------------------------
!! Returns the sum of inum independent gaussian noises squared, i.e. the result
!! is equivalent to summing the square of the return values of inum calls
!! to \(\texttt{gauss_dist}\).
!
IMPLICIT NONE
INTEGER, INTENT(IN) :: inum_gaussians
!
REAL(DP) sum_of_gaussians2
!
IF ( inum_gaussians < 0 ) THEN
CALL errore('sum_of_gaussians2', 'negative number of gaussians', 1)
ELSEIF ( inum_gaussians == 0 ) THEN
sum_of_gaussians2 = 0.0D0
ELSEIF ( inum_gaussians == 1 ) THEN
sum_of_gaussians2 = gauss_dist( 0.0D0, 1.0D0 )**2
ELSEIF( MODULO(inum_gaussians,2) == 0 ) THEN
sum_of_gaussians2 = 2.0 * gamma_dist( inum_gaussians/2 )
ELSE
sum_of_gaussians2 = 2.0 * gamma_dist((inum_gaussians-1)/2) + &
gauss_dist( 0.0D0, 1.0D0 )**2
ENDIF
!
ENDFUNCTION sum_of_gaussians2
!
END MODULE random_numbers