MODULE random ! A module for random number generation from the following distributions: ! ! Distribution Function/subroutine name ! ! Normal (Gaussian) random_normal ! Gamma random_gamma ! Chi-squared random_chisq ! Exponential random_exponential ! Weibull random_Weibull ! Beta random_beta ! t random_t ! Multivariate normal random_mvnorm ! Generalized inverse Gaussian random_inv_gauss ! Poisson random_Poisson ! Binomial random_binomial1 * ! random_binomial2 * ! Negative binomial random_neg_binomial ! von Mises random_von_Mises ! Cauchy random_Cauchy ! ! Generate a random ordering of the integers 1 .. N ! random_order ! Initialize (seed) the uniform random number generator for ANY compiler ! seed_random_number ! Lognormal - see note below. ! ** Two functions are provided for the binomial distribution. ! If the parameter values remain constant, it is recommended that the ! first function is used (random_binomial1). If one or both of the ! parameters change, use the second function (random_binomial2). ! The compilers own random number generator, SUBROUTINE RANDOM_NUMBER(r), ! is used to provide a source of uniformly distributed random numbers. ! N.B. At this stage, only one random number is generated at each call to ! one of the functions above. ! The module uses the following functions which are included here: ! bin_prob to calculate a single binomial probability ! lngamma to calculate the logarithm to base e of the gamma function ! Some of the code is adapted from Dagpunar's book: ! Dagpunar, J. 'Principles of random variate generation' ! Clarendon Press, Oxford, 1988. ISBN 0-19-852202-9 ! ! In most of Dagpunar's routines, there is a test to see whether the value ! of one or two floating-point parameters has changed since the last call. ! These tests have been replaced by using a logical variable FIRST. ! This should be set to .TRUE. on the first call using new values of the ! parameters, and .FALSE. if the parameter values are the same as for the ! previous call. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! Lognormal distribution ! If X has a lognormal distribution, then log(X) is normally distributed. ! Here the logarithm is the natural logarithm, that is to base e, sometimes ! denoted as ln. To generate random variates from this distribution, generate ! a random deviate from the normal distribution with mean and variance equal ! to the mean and variance of the logarithms of X, then take its exponential. ! Relationship between the mean & variance of log(X) and the mean & variance ! of X, when X has a lognormal distribution. ! Let m = mean of log(X), and s^2 = variance of log(X) ! Then ! mean of X = exp(m + 0.5s^2) ! variance of X = (mean(X))^2.[exp(s^2) - 1] ! In the reverse direction (rarely used) ! variance of log(X) = log[1 + var(X)/(mean(X))^2] ! mean of log(X) = log(mean(X) - 0.5var(log(X)) ! N.B. The above formulae relate to population parameters; they will only be ! approximate if applied to sample values. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! Version 1.13, 2 October 2000 ! Changes from version 1.01 ! 1. The random_order, random_Poisson & random_binomial routines have been ! replaced with more efficient routines. ! 2. A routine, seed_random_number, has been added to seed the uniform random ! number generator. This requires input of the required number of seeds ! for the particular compiler from a specified I/O unit such as a keyboard. ! 3. Made compatible with Lahey's ELF90. ! 4. Marsaglia & Tsang algorithm used for random_gamma when shape parameter > 1. ! 5. INTENT for array f corrected in random_mvnorm. ! Author: Alan Miller ! CSIRO Division of Mathematical & Information Sciences ! Private Bag 10, Clayton South MDC ! Clayton 3169, Victoria, Australia ! Phone: (+61) 3 9545-8016 Fax: (+61) 3 9545-8080 ! e-mail: amiller @ bigpond.net.au IMPLICIT NONE REAL, PRIVATE :: zero = 0.0, half = 0.5, one = 1.0, two = 2.0, & vsmall = TINY(1.0), vlarge = HUGE(1.0) PRIVATE :: integral INTEGER, PARAMETER :: dp = SELECTED_REAL_KIND(12, 60) CONTAINS FUNCTION random_normal() RESULT(fn_val) ! Adapted from the following Fortran 77 code ! ALGORITHM 712, COLLECTED ALGORITHMS FROM ACM. ! THIS WORK PUBLISHED IN TRANSACTIONS ON MATHEMATICAL SOFTWARE, ! VOL. 18, NO. 4, DECEMBER, 1992, PP. 434-435. ! The function random_normal() returns a normally distributed pseudo-random ! number with zero mean and unit variance. ! The algorithm uses the ratio of uniforms method of A.J. Kinderman ! and J.F. Monahan augmented with quadratic bounding curves. REAL :: fn_val ! Local variables REAL :: s = 0.449871, t = -0.386595, a = 0.19600, b = 0.25472, & r1 = 0.27597, r2 = 0.27846, u, v, x, y, q ! Generate P = (u,v) uniform in rectangle enclosing acceptance region DO CALL RANDOM_NUMBER(u) CALL RANDOM_NUMBER(v) v = 1.7156 * (v - half) ! Evaluate the quadratic form x = u - s y = ABS(v) - t q = x**2 + y*(a*y - b*x) ! Accept P if inside inner ellipse IF (q < r1) EXIT ! Reject P if outside outer ellipse IF (q > r2) CYCLE ! Reject P if outside acceptance region IF (v**2 < -4.0*LOG(u)*u**2) EXIT END DO ! Return ratio of P's coordinates as the normal deviate fn_val = v/u RETURN END FUNCTION random_normal FUNCTION random_gamma(s, first) RESULT(fn_val) ! Adapted from Fortran 77 code from the book: ! Dagpunar, J. 'Principles of random variate generation' ! Clarendon Press, Oxford, 1988. ISBN 0-19-852202-9 ! FUNCTION GENERATES A RANDOM GAMMA VARIATE. ! CALLS EITHER random_gamma1 (S > 1.0) ! OR random_exponential (S = 1.0) ! OR random_gamma2 (S < 1.0). ! S = SHAPE PARAMETER OF DISTRIBUTION (0 < REAL). REAL, INTENT(IN) :: s LOGICAL, INTENT(IN) :: first REAL :: fn_val IF (s <= zero) THEN WRITE(*, *) 'SHAPE PARAMETER VALUE MUST BE POSITIVE' STOP END IF IF (s > one) THEN fn_val = random_gamma1(s, first) ELSE IF (s < one) THEN fn_val = random_gamma2(s, first) ELSE fn_val = random_exponential() END IF RETURN END FUNCTION random_gamma FUNCTION random_gamma1(s, first) RESULT(fn_val) ! Uses the algorithm in ! Marsaglia, G. and Tsang, W.W. (2000) `A simple method for generating ! gamma variables', Trans. om Math. Software (TOMS), vol.26(3), pp.363-372. ! Generates a random gamma deviate for shape parameter s >= 1. REAL, INTENT(IN) :: s LOGICAL, INTENT(IN) :: first REAL :: fn_val ! Local variables REAL, SAVE :: c, d REAL :: u, v, x IF (first) THEN d = s - one/3. c = one/SQRT(9.0*d) END IF ! Start of main loop DO ! Generate v = (1+cx)^3 where x is random normal; repeat if v <= 0. DO x = random_normal() v = (one + c*x)**3 IF (v > zero) EXIT END DO ! Generate uniform variable U CALL RANDOM_NUMBER(u) IF (u < one - 0.0331*x**4) THEN fn_val = d*v EXIT ELSE IF (LOG(u) < half*x**2 + d*(one - v + LOG(v))) THEN fn_val = d*v EXIT END IF END DO RETURN END FUNCTION random_gamma1 FUNCTION random_gamma2(s, first) RESULT(fn_val) ! Adapted from Fortran 77 code from the book: ! Dagpunar, J. 'Principles of random variate generation' ! Clarendon Press, Oxford, 1988. ISBN 0-19-852202-9 ! FUNCTION GENERATES A RANDOM VARIATE IN [0,INFINITY) FROM ! A GAMMA DISTRIBUTION WITH DENSITY PROPORTIONAL TO ! GAMMA2**(S-1) * EXP(-GAMMA2), ! USING A SWITCHING METHOD. ! S = SHAPE PARAMETER OF DISTRIBUTION ! (REAL < 1.0) REAL, INTENT(IN) :: s LOGICAL, INTENT(IN) :: first REAL :: fn_val ! Local variables REAL :: r, x, w REAL, SAVE :: a, p, c, uf, vr, d IF (s <= zero .OR. s >= one) THEN WRITE(*, *) 'SHAPE PARAMETER VALUE OUTSIDE PERMITTED RANGE' STOP END IF IF (first) THEN ! Initialization, if necessary a = one - s p = a/(a + s*EXP(-a)) IF (s < vsmall) THEN WRITE(*, *) 'SHAPE PARAMETER VALUE TOO SMALL' STOP END IF c = one/s uf = p*(vsmall/a)**s vr = one - vsmall d = a*LOG(a) END IF DO CALL RANDOM_NUMBER(r) IF (r >= vr) THEN CYCLE ELSE IF (r > p) THEN x = a - LOG((one - r)/(one - p)) w = a*LOG(x)-d ELSE IF (r > uf) THEN x = a*(r/p)**c w = x ELSE fn_val = zero RETURN END IF CALL RANDOM_NUMBER(r) IF (one-r <= w .AND. r > zero) THEN IF (r*(w + one) >= one) CYCLE IF (-LOG(r) <= w) CYCLE END IF EXIT END DO fn_val = x RETURN END FUNCTION random_gamma2 FUNCTION random_chisq(ndf, first) RESULT(fn_val) ! Generates a random variate from the chi-squared distribution with ! ndf degrees of freedom INTEGER, INTENT(IN) :: ndf LOGICAL, INTENT(IN) :: first REAL :: fn_val fn_val = two * random_gamma(half*ndf, first) RETURN END FUNCTION random_chisq FUNCTION random_exponential() RESULT(fn_val) ! Adapted from Fortran 77 code from the book: ! Dagpunar, J. 'Principles of random variate generation' ! Clarendon Press, Oxford, 1988. ISBN 0-19-852202-9 ! FUNCTION GENERATES A RANDOM VARIATE IN [0,INFINITY) FROM ! A NEGATIVE EXPONENTIAL DlSTRIBUTION WlTH DENSITY PROPORTIONAL ! TO EXP(-random_exponential), USING INVERSION. REAL :: fn_val ! Local variable REAL :: r DO CALL RANDOM_NUMBER(r) IF (r > zero) EXIT END DO fn_val = -LOG(r) RETURN END FUNCTION random_exponential FUNCTION random_Weibull(a) RESULT(fn_val) ! Generates a random variate from the Weibull distribution with ! probability density: ! a ! a-1 -x ! f(x) = a.x e REAL, INTENT(IN) :: a REAL :: fn_val ! For speed, there is no checking that a is not zero or very small. fn_val = random_exponential() ** (one/a) RETURN END FUNCTION random_Weibull FUNCTION random_beta(aa, bb, first) RESULT(fn_val) ! Adapted from Fortran 77 code from the book: ! Dagpunar, J. 'Principles of random variate generation' ! Clarendon Press, Oxford, 1988. ISBN 0-19-852202-9 ! FUNCTION GENERATES A RANDOM VARIATE IN [0,1] ! FROM A BETA DISTRIBUTION WITH DENSITY ! PROPORTIONAL TO BETA**(AA-1) * (1-BETA)**(BB-1). ! USING CHENG'S LOG LOGISTIC METHOD. ! AA = SHAPE PARAMETER FROM DISTRIBUTION (0 < REAL) ! BB = SHAPE PARAMETER FROM DISTRIBUTION (0 < REAL) REAL, INTENT(IN) :: aa, bb LOGICAL, INTENT(IN) :: first REAL :: fn_val ! Local variables REAL, PARAMETER :: aln4 = 1.3862944 REAL :: a, b, g, r, s, x, y, z REAL, SAVE :: d, f, h, t, c LOGICAL, SAVE :: swap IF (aa <= zero .OR. bb <= zero) THEN WRITE(*, *) 'IMPERMISSIBLE SHAPE PARAMETER VALUE(S)' STOP END IF IF (first) THEN ! Initialization, if necessary a = aa b = bb swap = b > a IF (swap) THEN g = b b = a a = g END IF d = a/b f = a+b IF (b > one) THEN h = SQRT((two*a*b - f)/(f - two)) t = one ELSE h = b t = one/(one + (a/(vlarge*b))**b) END IF c = a+h END IF DO CALL RANDOM_NUMBER(r) CALL RANDOM_NUMBER(x) s = r*r*x IF (r < vsmall .OR. s <= zero) CYCLE IF (r < t) THEN x = LOG(r/(one - r))/h y = d*EXP(x) z = c*x + f*LOG((one + d)/(one + y)) - aln4 IF (s - one > z) THEN IF (s - s*z > one) CYCLE IF (LOG(s) > z) CYCLE END IF fn_val = y/(one + y) ELSE IF (4.0*s > (one + one/d)**f) CYCLE fn_val = one END IF EXIT END DO IF (swap) fn_val = one - fn_val RETURN END FUNCTION random_beta FUNCTION random_t(m) RESULT(fn_val) ! Adapted from Fortran 77 code from the book: ! Dagpunar, J. 'Principles of random variate generation' ! Clarendon Press, Oxford, 1988. ISBN 0-19-852202-9 ! FUNCTION GENERATES A RANDOM VARIATE FROM A ! T DISTRIBUTION USING KINDERMAN AND MONAHAN'S RATIO METHOD. ! M = DEGREES OF FREEDOM OF DISTRIBUTION ! (1 <= 1NTEGER) INTEGER, INTENT(IN) :: m REAL :: fn_val ! Local variables REAL, SAVE :: s, c, a, f, g REAL :: r, x, v REAL, PARAMETER :: three = 3.0, four = 4.0, quart = 0.25, & five = 5.0, sixteen = 16.0 INTEGER :: mm = 0 IF (m < 1) THEN WRITE(*, *) 'IMPERMISSIBLE DEGREES OF FREEDOM' STOP END IF IF (m /= mm) THEN ! Initialization, if necessary s = m c = -quart*(s + one) a = four/(one + one/s)**c f = sixteen/a IF (m > 1) THEN g = s - one g = ((s + one)/g)**c*SQRT((s+s)/g) ELSE g = one END IF mm = m END IF DO CALL RANDOM_NUMBER(r) IF (r <= zero) CYCLE CALL RANDOM_NUMBER(v) x = (two*v - one)*g/r v = x*x IF (v > five - a*r) THEN IF (m >= 1 .AND. r*(v + three) > f) CYCLE IF (r > (one + v/s)**c) CYCLE END IF EXIT END DO fn_val = x RETURN END FUNCTION random_t SUBROUTINE random_mvnorm(n, h, d, f, first, x, ier) ! Adapted from Fortran 77 code from the book: ! Dagpunar, J. 'Principles of random variate generation' ! Clarendon Press, Oxford, 1988. ISBN 0-19-852202-9 ! N.B. An extra argument, ier, has been added to Dagpunar's routine ! SUBROUTINE GENERATES AN N VARIATE RANDOM NORMAL ! VECTOR USING A CHOLESKY DECOMPOSITION. ! ARGUMENTS: ! N = NUMBER OF VARIATES IN VECTOR ! (INPUT,INTEGER >= 1) ! H(J) = J'TH ELEMENT OF VECTOR OF MEANS ! (INPUT,REAL) ! X(J) = J'TH ELEMENT OF DELIVERED VECTOR ! (OUTPUT,REAL) ! ! D(J*(J-1)/2+I) = (I,J)'TH ELEMENT OF VARIANCE MATRIX (J> = I) ! (INPUT,REAL) ! F((J-1)*(2*N-J)/2+I) = (I,J)'TH ELEMENT OF LOWER TRIANGULAR ! DECOMPOSITION OF VARIANCE MATRIX (J <= I) ! (OUTPUT,REAL) ! FIRST = .TRUE. IF THIS IS THE FIRST CALL OF THE ROUTINE ! OR IF THE DISTRIBUTION HAS CHANGED SINCE THE LAST CALL OF THE ROUTINE. ! OTHERWISE SET TO .FALSE. ! (INPUT,LOGICAL) ! ier = 1 if the input covariance matrix is not +ve definite ! = 0 otherwise INTEGER, INTENT(IN) :: n REAL, INTENT(IN) :: h(:), d(:) ! d(n*(n+1)/2) REAL, INTENT(IN OUT) :: f(:) ! f(n*(n+1)/2) REAL, INTENT(OUT) :: x(:) LOGICAL, INTENT(IN) :: first INTEGER, INTENT(OUT) :: ier ! Local variables INTEGER :: j, i, m REAL :: y, v INTEGER, SAVE :: n2 IF (n < 1) THEN WRITE(*, *) 'SIZE OF VECTOR IS NON POSITIVE' STOP END IF ier = 0 IF (first) THEN ! Initialization, if necessary n2 = 2*n IF (d(1) < zero) THEN ier = 1 RETURN END IF f(1) = SQRT(d(1)) y = one/f(1) DO j = 2,n f(j) = d(1+j*(j-1)/2) * y END DO DO i = 2,n v = d(i*(i-1)/2+i) DO m = 1,i-1 v = v - f((m-1)*(n2-m)/2+i)**2 END DO IF (v < zero) THEN ier = 1 RETURN END IF v = SQRT(v) y = one/v f((i-1)*(n2-i)/2+i) = v DO j = i+1,n v = d(j*(j-1)/2+i) DO m = 1,i-1 v = v - f((m-1)*(n2-m)/2+i)*f((m-1)*(n2-m)/2 + j) END DO ! m = 1,i-1 f((i-1)*(n2-i)/2 + j) = v*y END DO ! j = i+1,n END DO ! i = 2,n END IF x(1:n) = h(1:n) DO j = 1,n y = random_normal() DO i = j,n x(i) = x(i) + f((j-1)*(n2-j)/2 + i) * y END DO ! i = j,n END DO ! j = 1,n RETURN END SUBROUTINE random_mvnorm FUNCTION random_inv_gauss(h, b, first) RESULT(fn_val) ! Adapted from Fortran 77 code from the book: ! Dagpunar, J. 'Principles of random variate generation' ! Clarendon Press, Oxford, 1988. ISBN 0-19-852202-9 ! FUNCTION GENERATES A RANDOM VARIATE IN [0,INFINITY] FROM ! A REPARAMETERISED GENERALISED INVERSE GAUSSIAN (GIG) DISTRIBUTION ! WITH DENSITY PROPORTIONAL TO GIG**(H-1) * EXP(-0.5*B*(GIG+1/GIG)) ! USING A RATIO METHOD. ! H = PARAMETER OF DISTRIBUTION (0 <= REAL) ! B = PARAMETER OF DISTRIBUTION (0 < REAL) REAL, INTENT(IN) :: h, b LOGICAL, INTENT(IN) :: first REAL :: fn_val ! Local variables REAL :: ym, xm, r, w, r1, r2, x REAL, SAVE :: a, c, d, e REAL, PARAMETER :: quart = 0.25 IF (h < zero .OR. b <= zero) THEN WRITE(*, *) 'IMPERMISSIBLE DISTRIBUTION PARAMETER VALUES' STOP END IF IF (first) THEN ! Initialization, if necessary IF (h > quart*b*SQRT(vlarge)) THEN WRITE(*, *) 'THE RATIO H:B IS TOO SMALL' STOP END IF e = b*b d = h + one ym = (-d + SQRT(d*d + e))/b IF (ym < vsmall) THEN WRITE(*, *) 'THE VALUE OF B IS TOO SMALL' STOP END IF d = h - one xm = (d + SQRT(d*d + e))/b d = half*d e = -quart*b r = xm + one/xm w = xm*ym a = w**(-half*h) * SQRT(xm/ym) * EXP(-e*(r - ym - one/ym)) IF (a < vsmall) THEN WRITE(*, *) 'THE VALUE OF H IS TOO LARGE' STOP END IF c = -d*LOG(xm) - e*r END IF DO CALL RANDOM_NUMBER(r1) IF (r1 <= zero) CYCLE CALL RANDOM_NUMBER(r2) x = a*r2/r1 IF (x <= zero) CYCLE IF (LOG(r1) < d*LOG(x) + e*(x + one/x) + c) EXIT END DO fn_val = x RETURN END FUNCTION random_inv_gauss FUNCTION random_Poisson(mu, first) RESULT(ival) !********************************************************************** ! Translated to Fortran 90 by Alan Miller from: ! RANLIB ! ! Library of Fortran Routines for Random Number Generation ! ! Compiled and Written by: ! ! Barry W. Brown ! James Lovato ! ! Department of Biomathematics, Box 237 ! The University of Texas, M.D. Anderson Cancer Center ! 1515 Holcombe Boulevard ! Houston, TX 77030 ! ! This work was supported by grant CA-16672 from the National Cancer Institute. ! GENerate POIsson random deviate ! Function ! Generates a single random deviate from a Poisson distribution with mean mu. ! Arguments ! mu --> The mean of the Poisson distribution from which ! a random deviate is to be generated. ! REAL mu ! Method ! For details see: ! Ahrens, J.H. and Dieter, U. ! Computer Generation of Poisson Deviates ! From Modified Normal Distributions. ! ACM Trans. Math. Software, 8, 2 ! (June 1982),163-179 ! TABLES: COEFFICIENTS A0-A7 FOR STEP F. FACTORIALS FACT ! COEFFICIENTS A(K) - FOR PX = FK*V*V*SUM(A(K)*V**K)-DEL ! SEPARATION OF CASES A AND B ! .. Scalar Arguments .. REAL, INTENT(IN) :: mu LOGICAL, INTENT(IN) :: first INTEGER :: ival ! .. ! .. Local Scalars .. REAL :: b1, b2, c, c0, c1, c2, c3, del, difmuk, e, fk, fx, fy, g, & omega, px, py, t, u, v, x, xx REAL, SAVE :: s, d, p, q, p0 INTEGER :: j, k, kflag LOGICAL, SAVE :: full_init INTEGER, SAVE :: l, m ! .. ! .. Local Arrays .. REAL, SAVE :: pp(35) ! .. ! .. Data statements .. REAL, PARAMETER :: a0 = -.5, a1 = .3333333, a2 = -.2500068, a3 = .2000118, & a4 = -.1661269, a5 = .1421878, a6 = -.1384794, & a7 = .1250060 REAL, PARAMETER :: fact(10) = (/ 1., 1., 2., 6., 24., 120., 720., 5040., & 40320., 362880. /) ! .. ! .. Executable Statements .. IF (mu > 10.0) THEN ! C A S E A. (RECALCULATION OF S, D, L IF MU HAS CHANGED) IF (first) THEN s = SQRT(mu) d = 6.0*mu*mu ! THE POISSON PROBABILITIES PK EXCEED THE DISCRETE NORMAL ! PROBABILITIES FK WHENEVER K >= M(MU). L=IFIX(MU-1.1484) ! IS AN UPPER BOUND TO M(MU) FOR ALL MU >= 10 . l = mu - 1.1484 full_init = .false. END IF ! STEP N. NORMAL SAMPLE - random_normal() FOR STANDARD NORMAL DEVIATE g = mu + s*random_normal() IF (g > 0.0) THEN ival = g ! STEP I. IMMEDIATE ACCEPTANCE IF ival IS LARGE ENOUGH IF (ival>=l) RETURN ! STEP S. SQUEEZE ACCEPTANCE - SAMPLE U fk = ival difmuk = mu - fk CALL RANDOM_NUMBER(u) IF (d*u >= difmuk*difmuk*difmuk) RETURN END IF ! STEP P. PREPARATIONS FOR STEPS Q AND H. ! (RECALCULATIONS OF PARAMETERS IF NECESSARY) ! .3989423=(2*PI)**(-.5) .416667E-1=1./24. .1428571=1./7. ! THE QUANTITIES B1, B2, C3, C2, C1, C0 ARE FOR THE HERMITE ! APPROXIMATIONS TO THE DISCRETE NORMAL PROBABILITIES FK. ! C=.1069/MU GUARANTEES MAJORIZATION BY THE 'HAT'-FUNCTION. IF (.NOT. full_init) THEN omega = .3989423/s b1 = .4166667E-1/mu b2 = .3*b1*b1 c3 = .1428571*b1*b2 c2 = b2 - 15.*c3 c1 = b1 - 6.*b2 + 45.*c3 c0 = 1. - b1 + 3.*b2 - 15.*c3 c = .1069/mu full_init = .true. END IF IF (g < 0.0) GO TO 50 ! 'SUBROUTINE' F IS CALLED (KFLAG=0 FOR CORRECT RETURN) kflag = 0 GO TO 70 ! STEP Q. QUOTIENT ACCEPTANCE (RARE CASE) 40 IF (fy-u*fy <= py*EXP(px-fx)) RETURN ! STEP E. EXPONENTIAL SAMPLE - random_exponential() FOR STANDARD EXPONENTIAL ! DEVIATE E AND SAMPLE T FROM THE LAPLACE 'HAT' ! (IF T <= -.6744 THEN PK < FK FOR ALL MU >= 10.) 50 e = random_exponential() CALL RANDOM_NUMBER(u) u = u + u - one t = 1.8 + SIGN(e, u) IF (t <= (-.6744)) GO TO 50 ival = mu + s*t fk = ival difmuk = mu - fk ! 'SUBROUTINE' F IS CALLED (KFLAG=1 FOR CORRECT RETURN) kflag = 1 GO TO 70 ! STEP H. HAT ACCEPTANCE (E IS REPEATED ON REJECTION) 60 IF (c*ABS(u) > py*EXP(px+e) - fy*EXP(fx+e)) GO TO 50 RETURN ! STEP F. 'SUBROUTINE' F. CALCULATION OF PX, PY, FX, FY. ! CASE ival < 10 USES FACTORIALS FROM TABLE FACT 70 IF (ival>=10) GO TO 80 px = -mu py = mu**ival/fact(ival+1) GO TO 110 ! CASE ival >= 10 USES POLYNOMIAL APPROXIMATION ! A0-A7 FOR ACCURACY WHEN ADVISABLE ! .8333333E-1=1./12. .3989423=(2*PI)**(-.5) 80 del = .8333333E-1/fk del = del - 4.8*del*del*del v = difmuk/fk IF (ABS(v)>0.25) THEN px = fk*LOG(one + v) - difmuk - del ELSE px = fk*v*v* (((((((a7*v+a6)*v+a5)*v+a4)*v+a3)*v+a2)*v+a1)*v+a0) - del END IF py = .3989423/SQRT(fk) 110 x = (half - difmuk)/s xx = x*x fx = -half*xx fy = omega* (((c3*xx + c2)*xx + c1)*xx + c0) IF (kflag <= 0) GO TO 40 GO TO 60 !--------------------------------------------------------------------------- ! C A S E B. mu < 10 ! START NEW TABLE AND CALCULATE P0 IF NECESSARY ELSE IF (first) THEN m = MAX(1, INT(mu)) l = 0 p = EXP(-mu) q = p p0 = p END IF ! STEP U. UNIFORM SAMPLE FOR INVERSION METHOD DO CALL RANDOM_NUMBER(u) ival = 0 IF (u <= p0) RETURN ! STEP T. TABLE COMPARISON UNTIL THE END PP(L) OF THE ! PP-TABLE OF CUMULATIVE POISSON PROBABILITIES ! (0.458=PP(9) FOR MU=10) IF (l == 0) GO TO 150 j = 1 IF (u > 0.458) j = MIN(l, m) DO k = j, l IF (u <= pp(k)) GO TO 180 END DO IF (l == 35) CYCLE ! STEP C. CREATION OF NEW POISSON PROBABILITIES P ! AND THEIR CUMULATIVES Q=PP(K) 150 l = l + 1 DO k = l, 35 p = p*mu / k q = q + p pp(k) = q IF (u <= q) GO TO 170 END DO l = 35 END DO 170 l = k 180 ival = k RETURN END IF RETURN END FUNCTION random_Poisson FUNCTION random_binomial1(n, p, first) RESULT(ival) ! FUNCTION GENERATES A RANDOM BINOMIAL VARIATE USING C.D.Kemp's method. ! This algorithm is suitable when many random variates are required ! with the SAME parameter values for n & p. ! P = BERNOULLI SUCCESS PROBABILITY ! (0 <= REAL <= 1) ! N = NUMBER OF BERNOULLI TRIALS ! (1 <= INTEGER) ! FIRST = .TRUE. for the first call using the current parameter values ! = .FALSE. if the values of (n,p) are unchanged from last call ! Reference: Kemp, C.D. (1986). `A modal method for generating binomial ! variables', Commun. Statist. - Theor. Meth. 15(3), 805-813. INTEGER, INTENT(IN) :: n REAL, INTENT(IN) :: p LOGICAL, INTENT(IN) :: first INTEGER :: ival ! Local variables INTEGER :: ru, rd INTEGER, SAVE :: r0 REAL :: u, pd, pu REAL, SAVE :: odds_ratio, p_r REAL, PARAMETER :: zero = 0.0, one = 1.0 IF (first) THEN r0 = (n+1)*p p_r = bin_prob(n, p, r0) odds_ratio = p / (one - p) END IF CALL RANDOM_NUMBER(u) u = u - p_r IF (u < zero) THEN ival = r0 RETURN END IF pu = p_r ru = r0 pd = p_r rd = r0 DO rd = rd - 1 IF (rd >= 0) THEN pd = pd * (rd+1) / (odds_ratio * (n-rd)) u = u - pd IF (u < zero) THEN ival = rd RETURN END IF END IF ru = ru + 1 IF (ru <= n) THEN pu = pu * (n-ru+1) * odds_ratio / ru u = u - pu IF (u < zero) THEN ival = ru RETURN END IF END IF END DO ! This point should not be reached, but just in case: ival = r0 RETURN END FUNCTION random_binomial1 FUNCTION bin_prob(n, p, r) RESULT(fn_val) ! Calculate a binomial probability INTEGER, INTENT(IN) :: n, r REAL, INTENT(IN) :: p REAL :: fn_val ! Local variable REAL :: one = 1.0 fn_val = EXP( lngamma(DBLE(n+1)) - lngamma(DBLE(r+1)) - lngamma(DBLE(n-r+1)) & + r*LOG(p) + (n-r)*LOG(one - p) ) RETURN END FUNCTION bin_prob FUNCTION lngamma(x) RESULT(fn_val) ! Logarithm to base e of the gamma function. ! ! Accurate to about 1.e-14. ! Programmer: Alan Miller ! Latest revision of Fortran 77 version - 28 February 1988 REAL (dp), INTENT(IN) :: x REAL (dp) :: fn_val ! Local variables REAL (dp) :: a1 = -4.166666666554424D-02, a2 = 2.430554511376954D-03, & a3 = -7.685928044064347D-04, a4 = 5.660478426014386D-04, & temp, arg, product, lnrt2pi = 9.189385332046727D-1, & pi = 3.141592653589793D0 LOGICAL :: reflect ! lngamma is not defined if x = 0 or a negative integer. IF (x > 0.d0) GO TO 10 IF (x /= INT(x)) GO TO 10 fn_val = 0.d0 RETURN ! If x < 0, use the reflection formula: ! gamma(x) * gamma(1-x) = pi * cosec(pi.x) 10 reflect = (x < 0.d0) IF (reflect) THEN arg = 1.d0 - x ELSE arg = x END IF ! Increase the argument, if necessary, to make it > 10. product = 1.d0 20 IF (arg <= 10.d0) THEN product = product * arg arg = arg + 1.d0 GO TO 20 END IF ! Use a polynomial approximation to Stirling's formula. ! N.B. The real Stirling's formula is used here, not the simpler, but less ! accurate formula given by De Moivre in a letter to Stirling, which ! is the one usually quoted. arg = arg - 0.5D0 temp = 1.d0/arg**2 fn_val = lnrt2pi + arg * (LOG(arg) - 1.d0 + & (((a4*temp + a3)*temp + a2)*temp + a1)*temp) - LOG(product) IF (reflect) THEN temp = SIN(pi * x) fn_val = LOG(pi/temp) - fn_val END IF RETURN END FUNCTION lngamma FUNCTION random_binomial2(n, pp, first) RESULT(ival) !********************************************************************** ! Translated to Fortran 90 by Alan Miller from: ! RANLIB ! ! Library of Fortran Routines for Random Number Generation ! ! Compiled and Written by: ! ! Barry W. Brown ! James Lovato ! ! Department of Biomathematics, Box 237 ! The University of Texas, M.D. Anderson Cancer Center ! 1515 Holcombe Boulevard ! Houston, TX 77030 ! ! This work was supported by grant CA-16672 from the National Cancer Institute. ! GENerate BINomial random deviate ! Function ! Generates a single random deviate from a binomial ! distribution whose number of trials is N and whose ! probability of an event in each trial is P. ! Arguments ! N --> The number of trials in the binomial distribution ! from which a random deviate is to be generated. ! INTEGER N ! P --> The probability of an event in each trial of the ! binomial distribution from which a random deviate ! is to be generated. ! REAL P ! FIRST --> Set FIRST = .TRUE. for the first call to perform initialization ! the set FIRST = .FALSE. for further calls using the same pair ! of parameter values (N, P). ! LOGICAL FIRST ! random_binomial2 <-- A random deviate yielding the number of events ! from N independent trials, each of which has ! a probability of event P. ! INTEGER random_binomial ! Method ! This is algorithm BTPE from: ! Kachitvichyanukul, V. and Schmeiser, B. W. ! Binomial Random Variate Generation. ! Communications of the ACM, 31, 2 (February, 1988) 216. !********************************************************************** !*****DETERMINE APPROPRIATE ALGORITHM AND WHETHER SETUP IS NECESSARY ! .. ! .. Scalar Arguments .. REAL, INTENT(IN) :: pp INTEGER, INTENT(IN) :: n LOGICAL, INTENT(IN) :: first INTEGER :: ival ! .. ! .. Local Scalars .. REAL :: alv, amaxp, f, f1, f2, u, v, w, w2, x, x1, x2, ynorm, z, z2 REAL, PARAMETER :: zero = 0.0, half = 0.5, one = 1.0 INTEGER :: i, ix, ix1, k, mp INTEGER, SAVE :: m REAL, SAVE :: p, q, xnp, ffm, fm, xnpq, p1, xm, xl, xr, c, al, xll, & xlr, p2, p3, p4, qn, r, g ! .. ! .. Executable Statements .. !*****SETUP, PERFORM ONLY WHEN PARAMETERS CHANGE IF (first) THEN p = MIN(pp, one-pp) q = one - p xnp = n * p END IF IF (xnp > 30.) THEN IF (first) THEN ffm = xnp + p m = ffm fm = m xnpq = xnp * q p1 = INT(2.195*SQRT(xnpq) - 4.6*q) + half xm = fm + half xl = xm - p1 xr = xm + p1 c = 0.134 + 20.5 / (15.3 + fm) al = (ffm-xl) / (ffm - xl*p) xll = al * (one + half*al) al = (xr - ffm) / (xr*q) xlr = al * (one + half*al) p2 = p1 * (one + c + c) p3 = p2 + c / xll p4 = p3 + c / xlr END IF !*****GENERATE VARIATE, Binomial mean at least 30. 20 CALL RANDOM_NUMBER(u) u = u * p4 CALL RANDOM_NUMBER(v) ! TRIANGULAR REGION IF (u <= p1) THEN ix = xm - p1 * v + u GO TO 110 END IF ! PARALLELOGRAM REGION IF (u <= p2) THEN x = xl + (u-p1) / c v = v * c + one - ABS(xm-x) / p1 IF (v > one .OR. v <= zero) GO TO 20 ix = x ELSE ! LEFT TAIL IF (u <= p3) THEN ix = xl + LOG(v) / xll IF (ix < 0) GO TO 20 v = v * (u-p2) * xll ELSE ! RIGHT TAIL ix = xr - LOG(v) / xlr IF (ix > n) GO TO 20 v = v * (u-p3) * xlr END IF END IF !*****DETERMINE APPROPRIATE WAY TO PERFORM ACCEPT/REJECT TEST k = ABS(ix-m) IF (k <= 20 .OR. k >= xnpq/2-1) THEN ! EXPLICIT EVALUATION f = one r = p / q g = (n+1) * r IF (m < ix) THEN mp = m + 1 DO i = mp, ix f = f * (g/i-r) END DO ELSE IF (m > ix) THEN ix1 = ix + 1 DO i = ix1, m f = f / (g/i-r) END DO END IF IF (v > f) THEN GO TO 20 ELSE GO TO 110 END IF END IF ! SQUEEZING USING UPPER AND LOWER BOUNDS ON LOG(F(X)) amaxp = (k/xnpq) * ((k*(k/3. + .625) + .1666666666666)/xnpq + half) ynorm = -k * k / (2.*xnpq) alv = LOG(v) IF (alvynorm + amaxp) GO TO 20 ! STIRLING'S (actually de Moivre's) FORMULA TO MACHINE ACCURACY FOR ! THE FINAL ACCEPTANCE/REJECTION TEST x1 = ix + 1 f1 = fm + one z = n + 1 - fm w = n - ix + one z2 = z * z x2 = x1 * x1 f2 = f1 * f1 w2 = w * w IF (alv - (xm*LOG(f1/x1) + (n-m+half)*LOG(z/w) + (ix-m)*LOG(w*p/(x1*q)) + & (13860.-(462.-(132.-(99.-140./f2)/f2)/f2)/f2)/f1/166320. + & (13860.-(462.-(132.-(99.-140./z2)/z2)/z2)/z2)/z/166320. + & (13860.-(462.-(132.-(99.-140./x2)/x2)/x2)/x2)/x1/166320. + & (13860.-(462.-(132.-(99.-140./w2)/w2)/w2)/w2)/w/166320.) > zero) THEN GO TO 20 ELSE GO TO 110 END IF ELSE ! INVERSE CDF LOGIC FOR MEAN LESS THAN 30 IF (first) THEN qn = q ** n r = p / q g = r * (n+1) END IF 90 ix = 0 f = qn CALL RANDOM_NUMBER(u) 100 IF (u >= f) THEN IF (ix > 110) GO TO 90 u = u - f ix = ix + 1 f = f * (g/ix - r) GO TO 100 END IF END IF 110 IF (pp > half) ix = n - ix ival = ix RETURN END FUNCTION random_binomial2 FUNCTION random_neg_binomial(sk, p) RESULT(ival) ! Adapted from Fortran 77 code from the book: ! Dagpunar, J. 'Principles of random variate generation' ! Clarendon Press, Oxford, 1988. ISBN 0-19-852202-9 ! FUNCTION GENERATES A RANDOM NEGATIVE BINOMIAL VARIATE USING UNSTORED ! INVERSION AND/OR THE REPRODUCTIVE PROPERTY. ! SK = NUMBER OF FAILURES REQUIRED (Dagpunar's words!) ! = the `power' parameter of the negative binomial ! (0 < REAL) ! P = BERNOULLI SUCCESS PROBABILITY ! (0 < REAL < 1) ! THE PARAMETER H IS SET SO THAT UNSTORED INVERSION ONLY IS USED WHEN P <= H, ! OTHERWISE A COMBINATION OF UNSTORED INVERSION AND ! THE REPRODUCTIVE PROPERTY IS USED. REAL, INTENT(IN) :: sk, p INTEGER :: ival ! Local variables ! THE PARAMETER ULN = -LOG(MACHINE'S SMALLEST REAL NUMBER). REAL, PARAMETER :: h = 0.7 REAL :: q, x, st, uln, v, r, s, y, g INTEGER :: k, i, n IF (sk <= zero .OR. p <= zero .OR. p >= one) THEN WRITE(*, *) 'IMPERMISSIBLE DISTRIBUTION PARAMETER VALUES' STOP END IF q = one - p x = zero st = sk IF (p > h) THEN v = one/LOG(p) k = st DO i = 1,k DO CALL RANDOM_NUMBER(r) IF (r > zero) EXIT END DO n = v*LOG(r) x = x + n END DO st = st - k END IF s = zero uln = -LOG(vsmall) IF (st > -uln/LOG(q)) THEN WRITE(*, *) ' P IS TOO LARGE FOR THIS VALUE OF SK' STOP END IF y = q**st g = st CALL RANDOM_NUMBER(r) DO IF (y > r) EXIT r = r - y s = s + one y = y*p*g/s g = g + one END DO ival = x + s + half RETURN END FUNCTION random_neg_binomial FUNCTION random_von_Mises(k, first) RESULT(fn_val) ! Algorithm VMD from: ! Dagpunar, J.S. (1990) `Sampling from the von Mises distribution via a ! comparison of random numbers', J. of Appl. Statist., 17, 165-168. ! Fortran 90 code by Alan Miller ! CSIRO Division of Mathematical & Information Sciences ! Arguments: ! k (real) parameter of the von Mises distribution. ! first (logical) set to .TRUE. the first time that the function ! is called, or the first time with a new value ! for k. When first = .TRUE., the function sets ! up starting values and may be very much slower. REAL, INTENT(IN) :: k LOGICAL, INTENT(IN) :: first REAL :: fn_val ! Local variables INTEGER :: j, n INTEGER, SAVE :: nk REAL, PARAMETER :: pi = 3.14159265 REAL, SAVE :: p(20), theta(0:20) REAL :: sump, r, th, lambda, rlast REAL (dp) :: dk IF (first) THEN ! Initialization, if necessary IF (k < zero) THEN WRITE(*, *) '** Error: argument k for random_von_Mises = ', k RETURN END IF nk = k + k + one IF (nk > 20) THEN WRITE(*, *) '** Error: argument k for random_von_Mises = ', k RETURN END IF dk = k theta(0) = zero IF (k > half) THEN ! Set up array p of probabilities. sump = zero DO j = 1, nk IF (j < nk) THEN theta(j) = ACOS(one - j/k) ELSE theta(nk) = pi END IF ! Numerical integration of e^[k.cos(x)] from theta(j-1) to theta(j) CALL integral(theta(j-1), theta(j), p(j), dk) sump = sump + p(j) END DO p(1:nk) = p(1:nk) / sump ELSE p(1) = one theta(1) = pi END IF ! if k > 0.5 END IF ! if first CALL RANDOM_NUMBER(r) DO j = 1, nk r = r - p(j) IF (r < zero) EXIT END DO r = -r/p(j) DO th = theta(j-1) + r*(theta(j) - theta(j-1)) lambda = k - j + one - k*COS(th) n = 1 rlast = lambda DO CALL RANDOM_NUMBER(r) IF (r > rlast) EXIT n = n + 1 rlast = r END DO IF (n .NE. 2*(n/2)) EXIT ! is n even? CALL RANDOM_NUMBER(r) END DO fn_val = SIGN(th, (r - rlast)/(one - rlast) - half) RETURN END FUNCTION random_von_Mises SUBROUTINE integral(a, b, result, dk) ! Gaussian integration of exp(k.cosx) from a to b. REAL (dp), INTENT(IN) :: dk REAL, INTENT(IN) :: a, b REAL, INTENT(OUT) :: result ! Local variables REAL (dp) :: xmid, range, x1, x2, & x(3) = (/0.238619186083197_dp, 0.661209386466265_dp, 0.932469514203152_dp/), & w(3) = (/0.467913934572691_dp, 0.360761573048139_dp, 0.171324492379170_dp/) INTEGER :: i xmid = (a + b)/2._dp range = (b - a)/2._dp result = 0._dp DO i = 1, 3 x1 = xmid + x(i)*range x2 = xmid - x(i)*range result = result + w(i)*(EXP(dk*COS(x1)) + EXP(dk*COS(x2))) END DO result = result * range RETURN END SUBROUTINE integral FUNCTION random_Cauchy() RESULT(fn_val) ! Generate a random deviate from the standard Cauchy distribution REAL :: fn_val ! Local variables REAL :: v(2) DO CALL RANDOM_NUMBER(v) v = two*(v - half) IF (ABS(v(2)) < vsmall) CYCLE ! Test for zero IF (v(1)**2 + v(2)**2 < one) EXIT END DO fn_val = v(1) / v(2) RETURN END FUNCTION random_Cauchy SUBROUTINE random_order(order, n) ! Generate a random ordering of the integers 1 ... n. INTEGER, INTENT(IN) :: n INTEGER, INTENT(OUT) :: order(n) ! Local variables INTEGER :: i, j, k REAL :: wk DO i = 1, n order(i) = i END DO ! Starting at the end, swap the current last indicator with one ! randomly chosen from those preceeding it. DO i = n, 2, -1 CALL RANDOM_NUMBER(wk) j = 1 + i * wk IF (j < i) THEN k = order(i) order(i) = order(j) order(j) = k END IF END DO RETURN END SUBROUTINE random_order SUBROUTINE seed_random_number(iounit) INTEGER, INTENT(IN) :: iounit ! Local variables INTEGER :: k INTEGER, ALLOCATABLE :: seed(:) CALL RANDOM_SEED(SIZE=k) ALLOCATE( seed(k) ) WRITE(*, '(a, i2, a)')' Enter ', k, ' integers for random no. seeds: ' READ(*, *) seed WRITE(iounit, '(a, (7i10))') ' Random no. seeds: ', seed CALL RANDOM_SEED(PUT=seed) DEALLOCATE( seed ) RETURN END SUBROUTINE seed_random_number END MODULE random