Homec4science
Diffusion COSOlver (improved_sugama) Browse

cosolver/basis_transformation/MPI_T5/FM/FMsamples/SampleDPfunctions.f95
3c0a5500b5baimproved_sugama

Last Change
Raw File

SampleDPfunctions.f95
View Options

PROGRAM DP
! One use for FM involves programs that don't need multiple precision results but do need some
! of the special functions available in FM but not in the Fortran standard. These include:
! BERNOULLI(N)
! BETA(X,Y)
! BINOMIAL(N,K) or BINOMIAL(X,Y)
! COS_INTEGRAL(X)
! COSH_INTEGRAL(X)
! EXP_INTEGRAL_EI(X)
! EXP_INTEGRAL_EN(N,X)
! FRESNEL_C(X)
! FRESNEL_S(X)
! INCOMPLETE_BETA(X,A,B)
! INCOMPLETE_GAMMA1(X,Y)
! INCOMPLETE_GAMMA2(X,Y)
! LOG_INTEGRAL(X)
! POCHHAMMER(X,N)
! POLYGAMMA(N,X)
! PSI(X)
! SIN_INTEGRAL(X)
! SINH_INTEGRAL(X)
! See the complete list of FM functions in FM_User_Manual.txt.
! For this application, no TYPE(FM) variables need to be declared. Just add USE FMZM at the top
! and compile and link the program like SampleFM.f95.
USE FMZM
IMPLICIT NONE
INTEGER :: J
DOUBLE PRECISION :: A, B, C, C_FM, ERR, MAX_ERR
! To use with 53-bit double precision, having about 16 significant digits of accuracy,
! set the FM precision to 16 digits.
CALL FM_SET(16)
! 1. Check to see if Fortran's intrinsic gamma function is correctly rounded.
! A is the double precision variable, so GAMMA(A) uses Fortran's intrinsic gamma.
! TO_FM(A) converts A to an FM number, so GAMMA( TO_FM(A) ) uses FM's gamma,
! then the "=" rounds the result back to double precision variable C_FM.
! It is possible that different compilers might give different results for this
! test. Some compilers may not give results that are correctly rounded to full
! double precision accuray when A is large, but C_FM should be correctly rounded.
MAX_ERR = 0
DO J = 10, 150, 10
A = J + 0.5D0
C = GAMMA(A)
C_FM = GAMMA( TO_FM(A) )
ERR = ABS( (C - C_FM) / C_FM )
IF (ERR > MAX_ERR) THEN
MAX_ERR = ERR
B = A
ENDIF
ENDDO
WRITE (*,"(///A/)") " Sample 1. Compare Fortran's built-in gamma function to FM's"
IF (MAX_ERR > 0) THEN
A = B
WRITE (*,"(A,ES13.7,A,F7.3)") ' Maximum relative error in Fortran gamma was ', &
MAX_ERR, ' for A = ', A
C = GAMMA(A)
WRITE (*,"(ES25.15,A)") C, ' = GAMMA(A)'
C_FM = GAMMA( TO_FM(A) )
WRITE (*,"(ES25.15,A)") C_FM, ' = GAMMA( TO_FM(A) )'
ELSE
WRITE (*,"(A)") ' All Fortran gamma results were correctly rounded.'
ENDIF
! 2. Binomial coefficients.
! Find the probability of getting exactly 10,000 heads in 20,000 tosses
! of a fair coin.
! Here we could not store the results of the binomial and power separately in
! double precision, since BINOMIAL( 20000, 10000 ) = 2.2e+6018 and
! 2**20000 = 4.0e+6020 would both overflow in double precision.
WRITE (*,"(//A)") " Sample 2. Binomial coefficients"
WRITE (*,"(A)") " Find the probability of getting exactly 10,000 heads"
WRITE (*,"(A/)") " in 20,000 tosses of a fair coin."
C_FM = BINOMIAL( TO_FM(20000), TO_FM(10000) ) / TO_FM(2)**20000
WRITE (*,"(A,F20.16)") " BINOMIAL( TO_FM(20000), TO_FM(10000) ) / TO_FM(2)**20000 =", C_FM
! 3. Log Integral function.
! Estimate the number of primes less than 10**30.
WRITE (*,"(//A)") " Sample 3. Log integral"
WRITE (*,"(A/)") " Estimate the number of primes less than 10**30."
C_FM = LOG_INTEGRAL( TO_FM('1.0E+30') )
WRITE (*,"(A,ES23.15)") " LOG_INTEGRAL(TO_FM('1.0E+30')) =", C_FM
! 4. Psi and polygamma functions.
! Rational series can often be summed using these functions.
! Sum (n=1 to infinity) 1/(n**2 * (8n+1)**2) =
! 16*(psi(1) - psi(9/8)) + polygamma(1,1) + polygamma(1,9/8)
! Reference: Abramowitz & Stegun, Handbook of Mathematical Functions,
! chapter 6, Example 10.
WRITE (*,"(//A)") " Sample 4. Psi and polygamma functions."
WRITE (*,"(A)") " Sum (n=1 to infinity) 1/(n**2 * (8n+1)**2) ="
WRITE (*,"(A/)") " 16*(psi(1) - psi(9/8)) + polygamma(1,1) + polygamma(1,9/8)"
C_FM = 16*( PSI( TO_FM(1) ) - PSI( TO_FM(9)/8 ) ) + &
POLYGAMMA( 1, TO_FM(1) ) + POLYGAMMA( 1, TO_FM(9)/8 )
WRITE (*,"(A,F19.16)") " Sum =", C_FM
! 5. Incomplete gamma and gamma functions.
! Find the probability that an observed chi-square for a correct model should be
! less that 2.3 when the number of degrees of freedom is 5.
! Reference: Knuth, Volume 2, 3rd ed., Page 56, and Press, Flannery, Teukolsky,
! Vetterling, Numerical Recipes, 1st ed., Page 165.
WRITE (*,"(//A/)") " Sample 5. Incomplete gamma and gamma functions."
C_FM = INCOMPLETE_GAMMA1( TO_FM(5)/2, TO_FM('2.3')/2 ) / GAMMA( TO_FM(5)/2 )
WRITE (*,"(A,F19.16/)") " Probability =", C_FM
END PROGRAM DP
c4science · Help