2019-01-25 11:39:31 +01:00
|
|
|
subroutine give_explicit_poly_and_gaussian_x(P_new,P_center,p,fact_k,iorder,alpha,beta,a,b,A_center,B_center,dim)
|
|
|
|
BEGIN_DOC
|
|
|
|
! Transform the product of
|
|
|
|
! (x-x_A)^a(1) (x-x_B)^b(1) (x-x_A)^a(2) (y-y_B)^b(2) (z-z_A)^a(3) (z-z_B)^b(3) exp(-(r-A)^2 alpha) exp(-(r-B)^2 beta)
|
|
|
|
! into
|
|
|
|
! fact_k (x-x_P)^iorder(1) (y-y_P)^iorder(2) (z-z_P)^iorder(3) exp(-p(r-P)^2)
|
|
|
|
END_DOC
|
|
|
|
implicit none
|
|
|
|
include 'constants.include.F'
|
|
|
|
integer, intent(in) :: dim
|
|
|
|
integer, intent(in) :: a,b ! powers : (x-xa)**a_x = (x-A(1))**a(1)
|
|
|
|
double precision, intent(in) :: alpha, beta ! exponents
|
|
|
|
double precision, intent(in) :: A_center ! A center
|
|
|
|
double precision, intent(in) :: B_center ! B center
|
|
|
|
double precision, intent(out) :: P_center ! new center
|
|
|
|
double precision, intent(out) :: p ! new exponent
|
|
|
|
double precision, intent(out) :: fact_k ! constant factor
|
|
|
|
double precision, intent(out) :: P_new(0:max_dim) ! polynomial
|
|
|
|
integer, intent(out) :: iorder ! order of the polynomials
|
|
|
|
|
|
|
|
double precision :: P_a(0:max_dim), P_b(0:max_dim)
|
|
|
|
integer :: n_new,i,j
|
|
|
|
double precision :: p_inv,ab,d_AB
|
|
|
|
!DIR$ ATTRIBUTES ALIGN : $IRP_ALIGN :: P_a, P_b
|
|
|
|
|
|
|
|
! Do the gaussian product to get the new center and the new exponent
|
|
|
|
P_new = 0.d0
|
|
|
|
p = alpha+beta
|
|
|
|
p_inv = 1.d0/p
|
|
|
|
ab = alpha * beta
|
|
|
|
d_AB = (A_center - B_center) * (A_center - B_center)
|
|
|
|
P_center = (alpha * A_center + beta * B_center) * p_inv
|
2020-08-07 11:47:51 +02:00
|
|
|
if(dabs(ab*p_inv * d_AB).lt.50.d0)then
|
|
|
|
fact_k = exp(-ab*p_inv * d_AB)
|
|
|
|
else
|
|
|
|
fact_k = 0.d0
|
|
|
|
endif
|
2019-01-25 11:39:31 +01:00
|
|
|
|
|
|
|
! Recenter the polynomials P_a and P_b on x
|
|
|
|
!DIR$ FORCEINLINE
|
|
|
|
call recentered_poly2(P_a(0),A_center,P_center,a,P_b(0),B_center,P_center,b)
|
|
|
|
n_new = 0
|
|
|
|
|
|
|
|
!DIR$ FORCEINLINE
|
|
|
|
call multiply_poly(P_a(0),a,P_b(0),b,P_new(0),n_new)
|
|
|
|
iorder = a + b
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
|
|
subroutine give_explicit_poly_and_gaussian(P_new,P_center,p,fact_k,iorder,alpha,beta,a,b,A_center,B_center,dim)
|
|
|
|
BEGIN_DOC
|
|
|
|
! Transforms the product of
|
|
|
|
! (x-x_A)^a(1) (x-x_B)^b(1) (x-x_A)^a(2) (y-y_B)^b(2) (z-z_A)^a(3) (z-z_B)^b(3) exp(-(r-A)^2 alpha) exp(-(r-B)^2 beta)
|
|
|
|
! into
|
|
|
|
! fact_k * [ sum (l_x = 0,i_order(1)) P_new(l_x,1) * (x-P_center(1))^l_x ] exp (- p (x-P_center(1))^2 )
|
|
|
|
! * [ sum (l_y = 0,i_order(2)) P_new(l_y,2) * (y-P_center(2))^l_y ] exp (- p (y-P_center(2))^2 )
|
|
|
|
! * [ sum (l_z = 0,i_order(3)) P_new(l_z,3) * (z-P_center(3))^l_z ] exp (- p (z-P_center(3))^2 )
|
2021-06-28 11:52:06 +02:00
|
|
|
!
|
|
|
|
! WARNING ::: IF fact_k is too smal then:
|
|
|
|
! returns a "s" function centered in zero
|
|
|
|
! with an inifinite exponent and a zero polynom coef
|
2019-01-25 11:39:31 +01:00
|
|
|
END_DOC
|
|
|
|
implicit none
|
|
|
|
include 'constants.include.F'
|
|
|
|
integer, intent(in) :: dim
|
|
|
|
integer, intent(in) :: a(3),b(3) ! powers : (x-xa)**a_x = (x-A(1))**a(1)
|
|
|
|
double precision, intent(in) :: alpha, beta ! exponents
|
|
|
|
double precision, intent(in) :: A_center(3) ! A center
|
|
|
|
double precision, intent(in) :: B_center (3) ! B center
|
|
|
|
double precision, intent(out) :: P_center(3) ! new center
|
|
|
|
double precision, intent(out) :: p ! new exponent
|
|
|
|
double precision, intent(out) :: fact_k ! constant factor
|
|
|
|
double precision, intent(out) :: P_new(0:max_dim,3)! polynomial
|
|
|
|
integer, intent(out) :: iorder(3) ! i_order(i) = order of the polynomials
|
|
|
|
|
|
|
|
double precision :: P_a(0:max_dim,3), P_b(0:max_dim,3)
|
|
|
|
integer :: n_new,i,j
|
|
|
|
!DIR$ ATTRIBUTES ALIGN : $IRP_ALIGN :: P_a, P_b
|
|
|
|
|
|
|
|
iorder(1) = 0
|
|
|
|
iorder(2) = 0
|
|
|
|
iorder(3) = 0
|
|
|
|
P_new(0,1) = 0.d0
|
|
|
|
P_new(0,2) = 0.d0
|
|
|
|
P_new(0,3) = 0.d0
|
|
|
|
!DIR$ FORCEINLINE
|
|
|
|
call gaussian_product(alpha,A_center,beta,B_center,fact_k,p,P_center)
|
|
|
|
if (fact_k < thresh) then
|
2021-06-28 11:52:06 +02:00
|
|
|
! IF fact_k is too smal then:
|
|
|
|
! returns a "s" function centered in zero
|
|
|
|
! with an inifinite exponent and a zero polynom coef
|
2020-08-07 11:47:51 +02:00
|
|
|
P_center = 0.d0
|
2021-06-28 11:52:06 +02:00
|
|
|
p = 1.d+15
|
2020-08-07 11:47:51 +02:00
|
|
|
P_new = 0.d0
|
2021-06-28 11:52:06 +02:00
|
|
|
iorder = 0
|
2019-01-25 11:39:31 +01:00
|
|
|
fact_k = 0.d0
|
|
|
|
return
|
|
|
|
endif
|
|
|
|
|
|
|
|
!DIR$ FORCEINLINE
|
|
|
|
call recentered_poly2(P_a(0,1),A_center(1),P_center(1),a(1),P_b(0,1),B_center(1),P_center(1),b(1))
|
|
|
|
iorder(1) = a(1) + b(1)
|
|
|
|
do i=0,iorder(1)
|
|
|
|
P_new(i,1) = 0.d0
|
|
|
|
enddo
|
|
|
|
n_new=0
|
|
|
|
!DIR$ FORCEINLINE
|
|
|
|
call multiply_poly(P_a(0,1),a(1),P_b(0,1),b(1),P_new(0,1),n_new)
|
|
|
|
|
|
|
|
!DIR$ FORCEINLINE
|
|
|
|
call recentered_poly2(P_a(0,2),A_center(2),P_center(2),a(2),P_b(0,2),B_center(2),P_center(2),b(2))
|
|
|
|
iorder(2) = a(2) + b(2)
|
|
|
|
do i=0,iorder(2)
|
|
|
|
P_new(i,2) = 0.d0
|
|
|
|
enddo
|
|
|
|
n_new=0
|
|
|
|
!DIR$ FORCEINLINE
|
|
|
|
call multiply_poly(P_a(0,2),a(2),P_b(0,2),b(2),P_new(0,2),n_new)
|
|
|
|
|
|
|
|
!DIR$ FORCEINLINE
|
|
|
|
call recentered_poly2(P_a(0,3),A_center(3),P_center(3),a(3),P_b(0,3),B_center(3),P_center(3),b(3))
|
|
|
|
iorder(3) = a(3) + b(3)
|
|
|
|
do i=0,iorder(3)
|
|
|
|
P_new(i,3) = 0.d0
|
|
|
|
enddo
|
|
|
|
n_new=0
|
|
|
|
!DIR$ FORCEINLINE
|
|
|
|
call multiply_poly(P_a(0,3),a(3),P_b(0,3),b(3),P_new(0,3),n_new)
|
|
|
|
|
|
|
|
end
|
|
|
|
|
2023-02-06 19:00:35 +01:00
|
|
|
subroutine give_explicit_poly_and_gaussian_v(P_new, ldp, P_center, p, fact_k, iorder, alpha, beta, a, b, A_center, LD_A, B_center, n_points)
|
|
|
|
|
|
|
|
BEGIN_DOC
|
|
|
|
! Transforms the product of
|
2023-03-04 17:49:48 +01:00
|
|
|
! (x-x_A)^a(1) (x-x_B)^b(1) (y-y_A)^a(2) (y-y_B)^b(2) (z-z_A)^a(3) (z-z_B)^b(3) exp(-(r-A)^2 alpha) exp(-(r-B)^2 beta)
|
2023-02-06 19:00:35 +01:00
|
|
|
! into
|
|
|
|
! fact_k * [ sum (l_x = 0,i_order(1)) P_new(l_x,1) * (x-P_center(1))^l_x ] exp (- p (x-P_center(1))^2 )
|
|
|
|
! * [ sum (l_y = 0,i_order(2)) P_new(l_y,2) * (y-P_center(2))^l_y ] exp (- p (y-P_center(2))^2 )
|
|
|
|
! * [ sum (l_z = 0,i_order(3)) P_new(l_z,3) * (z-P_center(3))^l_z ] exp (- p (z-P_center(3))^2 )
|
|
|
|
!
|
|
|
|
! WARNING :: : IF fact_k is too smal then:
|
|
|
|
! returns a "s" function centered in zero
|
|
|
|
! with an inifinite exponent and a zero polynom coef
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
include 'constants.include.F'
|
|
|
|
|
|
|
|
implicit none
|
|
|
|
integer, intent(in) :: n_points, ldp, LD_A
|
|
|
|
integer, intent(in) :: a(3), b(3) ! powers : (x-xa)**a_x = (x-A(1))**a(1)
|
|
|
|
double precision, intent(in) :: alpha, beta ! exponents
|
|
|
|
double precision, intent(in) :: A_center(LD_A,3) ! A center
|
|
|
|
double precision, intent(in) :: B_center(3) ! B center
|
|
|
|
integer, intent(out) :: iorder(3) ! i_order(i) = order of the polynomials
|
|
|
|
double precision, intent(out) :: P_center(n_points,3) ! new center
|
|
|
|
double precision, intent(out) :: p ! new exponent
|
|
|
|
double precision, intent(out) :: fact_k(n_points) ! constant factor
|
|
|
|
double precision, intent(out) :: P_new(n_points,0:ldp,3) ! polynomial
|
|
|
|
|
|
|
|
integer :: n_new, i, j, ipoint, lda, ldb, xyz
|
|
|
|
double precision, allocatable :: P_a(:,:,:), P_b(:,:,:)
|
|
|
|
|
|
|
|
|
|
|
|
call gaussian_product_v(alpha, A_center, LD_A, beta, B_center, fact_k, p, P_center, n_points)
|
|
|
|
|
|
|
|
if(ior(ior(b(1), b(2)), b(3)) == 0) then ! b == (0,0,0)
|
|
|
|
|
|
|
|
iorder(1:3) = a(1:3)
|
|
|
|
|
|
|
|
lda = maxval(a)
|
|
|
|
allocate(P_a(n_points,0:lda,3))
|
|
|
|
!ldb = 0
|
|
|
|
!allocate(P_b(n_points,0:0,3))
|
|
|
|
|
|
|
|
!call recentered_poly2_v0(P_a, lda, A_center, LD_A, P_center, a, P_b, B_center, P_center, n_points)
|
|
|
|
call recentered_poly2_v0(P_a, lda, A_center, LD_A, P_center, a, n_points)
|
|
|
|
|
|
|
|
do ipoint = 1, n_points
|
|
|
|
do xyz = 1, 3
|
|
|
|
!P_new(ipoint,0,xyz) = P_a(ipoint,0,xyz) * P_b(ipoint,0,xyz)
|
|
|
|
P_new(ipoint,0,xyz) = P_a(ipoint,0,xyz)
|
|
|
|
do i = 1, a(xyz)
|
|
|
|
!P_new(ipoint,i,xyz) = P_new(ipoint,i,xyz) + P_b(ipoint,0,xyz) * P_a(ipoint,i,xyz)
|
|
|
|
P_new(ipoint,i,xyz) = P_a(ipoint,i,xyz)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
|
|
|
|
deallocate(P_a)
|
|
|
|
!deallocate(P_b)
|
|
|
|
|
|
|
|
return
|
|
|
|
endif
|
|
|
|
|
|
|
|
lda = maxval(a)
|
|
|
|
ldb = maxval(b)
|
|
|
|
allocate(P_a(n_points,0:lda,3), P_b(n_points,0:ldb,3))
|
|
|
|
|
|
|
|
call recentered_poly2_v(P_a, lda, A_center, LD_A, P_center, a, P_b, ldb, B_center, P_center, b, n_points)
|
|
|
|
|
|
|
|
iorder(1:3) = a(1:3) + b(1:3)
|
|
|
|
|
|
|
|
do xyz = 1, 3
|
|
|
|
if(b(xyz) == 0) then
|
|
|
|
|
|
|
|
do ipoint = 1, n_points
|
|
|
|
!P_new(ipoint,0,xyz) = P_a(ipoint,0,xyz) * P_b(ipoint,0,xyz)
|
|
|
|
P_new(ipoint,0,xyz) = P_a(ipoint,0,xyz)
|
|
|
|
do i = 1, a(xyz)
|
|
|
|
!P_new(ipoint,i,xyz) = P_new(ipoint,i,xyz) + P_b(ipoint,0,xyz) * P_a(ipoint,i,xyz)
|
|
|
|
P_new(ipoint,i,xyz) = P_a(ipoint,i,xyz)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
|
|
|
|
else
|
|
|
|
|
|
|
|
do i = 0, iorder(xyz)
|
|
|
|
do ipoint = 1, n_points
|
|
|
|
P_new(ipoint,i,xyz) = 0.d0
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
|
|
|
|
call multiply_poly_v(P_a(1,0,xyz), a(xyz), P_b(1,0,xyz), b(xyz), P_new(1,0,xyz), ldp, n_points)
|
|
|
|
|
|
|
|
endif
|
|
|
|
enddo
|
|
|
|
|
|
|
|
end subroutine give_explicit_poly_and_gaussian_v
|
|
|
|
|
|
|
|
! ---
|
2019-01-25 11:39:31 +01:00
|
|
|
|
|
|
|
subroutine give_explicit_poly_and_gaussian_double(P_new,P_center,p,fact_k,iorder,alpha,beta,gama,a,b,A_center,B_center,Nucl_center,dim)
|
|
|
|
BEGIN_DOC
|
|
|
|
! Transforms the product of
|
|
|
|
! (x-x_A)^a(1) (x-x_B)^b(1) (x-x_A)^a(2) (y-y_B)^b(2) (z-z_A)^a(3) (z-z_B)^b(3)
|
|
|
|
! exp(-(r-A)^2 alpha) exp(-(r-B)^2 beta) exp(-(r-Nucl_center)^2 gama
|
|
|
|
!
|
|
|
|
! into
|
|
|
|
! fact_k * [ sum (l_x = 0,i_order(1)) P_new(l_x,1) * (x-P_center(1))^l_x ] exp (- p (x-P_center(1))^2 )
|
|
|
|
! * [ sum (l_y = 0,i_order(2)) P_new(l_y,2) * (y-P_center(2))^l_y ] exp (- p (y-P_center(2))^2 )
|
|
|
|
! * [ sum (l_z = 0,i_order(3)) P_new(l_z,3) * (z-P_center(3))^l_z ] exp (- p (z-P_center(3))^2 )
|
|
|
|
END_DOC
|
|
|
|
implicit none
|
|
|
|
include 'constants.include.F'
|
|
|
|
integer, intent(in) :: dim
|
|
|
|
integer, intent(in) :: a(3),b(3) ! powers : (x-xa)**a_x = (x-A(1))**a(1)
|
|
|
|
double precision, intent(in) :: alpha, beta, gama ! exponents
|
|
|
|
double precision, intent(in) :: A_center(3) ! A center
|
|
|
|
double precision, intent(in) :: B_center (3) ! B center
|
|
|
|
double precision, intent(in) :: Nucl_center(3) ! B center
|
|
|
|
double precision, intent(out) :: P_center(3) ! new center
|
|
|
|
double precision, intent(out) :: p ! new exponent
|
|
|
|
double precision, intent(out) :: fact_k ! constant factor
|
|
|
|
double precision, intent(out) :: P_new(0:max_dim,3)! polynomial
|
|
|
|
integer , intent(out) :: iorder(3) ! i_order(i) = order of the polynomials
|
|
|
|
|
|
|
|
double precision :: P_center_tmp(3) ! new center
|
|
|
|
double precision :: p_tmp ! new exponent
|
|
|
|
double precision :: fact_k_tmp,fact_k_bis ! constant factor
|
|
|
|
double precision :: P_new_tmp(0:max_dim,3)! polynomial
|
|
|
|
integer :: i,j
|
|
|
|
double precision :: binom_func
|
|
|
|
|
|
|
|
! First you transform the two primitives into a sum of primitive with the same center P_center_tmp and gaussian exponent p_tmp
|
|
|
|
call give_explicit_poly_and_gaussian(P_new_tmp,P_center_tmp,p_tmp,fact_k_tmp,iorder,alpha,beta,a,b,A_center,B_center,dim)
|
|
|
|
! Then you create the new gaussian from the product of the new one per the Nuclei one
|
|
|
|
call gaussian_product(p_tmp,P_center_tmp,gama,Nucl_center,fact_k_bis,p,P_center)
|
|
|
|
fact_k = fact_k_bis * fact_k_tmp
|
|
|
|
|
|
|
|
! Then you build the coefficient of the new polynom
|
|
|
|
do i = 0, iorder(1)
|
|
|
|
P_new(i,1) = 0.d0
|
|
|
|
do j = i,iorder(1)
|
|
|
|
P_new(i,1) = P_new(i,1) + P_new_tmp(j,1) * binom_func(j,j-i) * (P_center(1) - P_center_tmp(1))**(j-i)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
do i = 0, iorder(2)
|
|
|
|
P_new(i,2) = 0.d0
|
|
|
|
do j = i,iorder(2)
|
|
|
|
P_new(i,2) = P_new(i,2) + P_new_tmp(j,2) * binom_func(j,j-i) * (P_center(2) - P_center_tmp(2))**(j-i)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
do i = 0, iorder(3)
|
|
|
|
P_new(i,3) = 0.d0
|
|
|
|
do j = i,iorder(3)
|
|
|
|
P_new(i,3) = P_new(i,3) + P_new_tmp(j,3) * binom_func(j,j-i) * (P_center(3) - P_center_tmp(3))**(j-i)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
subroutine gaussian_product(a,xa,b,xb,k,p,xp)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Gaussian product in 1D.
|
|
|
|
! e^{-a (x-x_A)^2} e^{-b (x-x_B)^2} = K_{ab}^x e^{-p (x-x_P)^2}
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
double precision, intent(in) :: a,b ! Exponents
|
|
|
|
double precision, intent(in) :: xa(3),xb(3) ! Centers
|
|
|
|
double precision, intent(out) :: p ! New exponent
|
|
|
|
double precision, intent(out) :: xp(3) ! New center
|
|
|
|
double precision, intent(out) :: k ! Constant
|
|
|
|
|
|
|
|
double precision :: p_inv
|
|
|
|
|
|
|
|
ASSERT (a>0.)
|
|
|
|
ASSERT (b>0.)
|
|
|
|
|
|
|
|
double precision :: xab(3), ab
|
|
|
|
!DIR$ ATTRIBUTES ALIGN : $IRP_ALIGN :: xab
|
|
|
|
|
|
|
|
p = a+b
|
|
|
|
p_inv = 1.d0/(a+b)
|
|
|
|
ab = a*b
|
|
|
|
xab(1) = xa(1)-xb(1)
|
|
|
|
xab(2) = xa(2)-xb(2)
|
|
|
|
xab(3) = xa(3)-xb(3)
|
|
|
|
ab = ab*p_inv
|
|
|
|
k = ab*(xab(1)*xab(1)+xab(2)*xab(2)+xab(3)*xab(3))
|
|
|
|
if (k > 40.d0) then
|
|
|
|
k=0.d0
|
|
|
|
return
|
|
|
|
endif
|
|
|
|
k = dexp(-k)
|
|
|
|
xp(1) = (a*xa(1)+b*xb(1))*p_inv
|
|
|
|
xp(2) = (a*xa(2)+b*xb(2))*p_inv
|
|
|
|
xp(3) = (a*xa(3)+b*xb(3))*p_inv
|
|
|
|
end subroutine
|
|
|
|
|
|
|
|
|
2023-02-06 19:00:35 +01:00
|
|
|
subroutine gaussian_product_v(a, xa, LD_xa, b, xb, k, p, xp, n_points)
|
|
|
|
|
|
|
|
BEGIN_DOC
|
|
|
|
!
|
|
|
|
! Gaussian product in 1D.
|
|
|
|
! e^{-a (x-x_A)^2} e^{-b (x-x_B)^2} = K_{ab}^x e^{-p (x-x_P)^2}
|
|
|
|
!
|
|
|
|
! Using multiple A centers
|
|
|
|
!
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
implicit none
|
|
|
|
|
|
|
|
integer, intent(in) :: LD_xa, n_points
|
|
|
|
double precision, intent(in) :: a, b ! Exponents
|
|
|
|
double precision, intent(in) :: xa(LD_xa,3), xb(3) ! Centers
|
|
|
|
double precision, intent(out) :: p ! New exponent
|
|
|
|
double precision, intent(out) :: xp(n_points,3) ! New center
|
|
|
|
double precision, intent(out) :: k(n_points) ! Constant
|
|
|
|
|
|
|
|
integer :: ipoint
|
|
|
|
double precision :: p_inv
|
|
|
|
double precision :: xab(3), ab, ap, bp, bpxb(3)
|
|
|
|
!DIR$ ATTRIBUTES ALIGN : $IRP_ALIGN :: xab
|
|
|
|
|
|
|
|
ASSERT (a>0.)
|
|
|
|
ASSERT (b>0.)
|
|
|
|
|
|
|
|
p = a+b
|
|
|
|
p_inv = 1.d0/(a+b)
|
|
|
|
ab = a*b*p_inv
|
|
|
|
ap = a*p_inv
|
|
|
|
bp = b*p_inv
|
|
|
|
bpxb(1) = bp*xb(1)
|
|
|
|
bpxb(2) = bp*xb(2)
|
|
|
|
bpxb(3) = bp*xb(3)
|
|
|
|
|
|
|
|
do ipoint = 1, n_points
|
|
|
|
xab(1) = xa(ipoint,1)-xb(1)
|
|
|
|
xab(2) = xa(ipoint,2)-xb(2)
|
|
|
|
xab(3) = xa(ipoint,3)-xb(3)
|
|
|
|
k(ipoint) = ab*(xab(1)*xab(1)+xab(2)*xab(2)+xab(3)*xab(3))
|
|
|
|
if (k(ipoint) > 40.d0) then
|
|
|
|
k(ipoint)=0.d0
|
|
|
|
xp(ipoint,1) = 0.d0
|
|
|
|
xp(ipoint,2) = 0.d0
|
|
|
|
xp(ipoint,3) = 0.d0
|
|
|
|
else
|
|
|
|
k(ipoint) = dexp(-k(ipoint))
|
|
|
|
xp(ipoint,1) = ap*xa(ipoint,1)+bpxb(1)
|
|
|
|
xp(ipoint,2) = ap*xa(ipoint,2)+bpxb(2)
|
|
|
|
xp(ipoint,3) = ap*xa(ipoint,3)+bpxb(3)
|
|
|
|
endif
|
|
|
|
enddo
|
|
|
|
|
|
|
|
end subroutine gaussian_product_v
|
|
|
|
|
|
|
|
! ---
|
2019-01-25 11:39:31 +01:00
|
|
|
|
|
|
|
|
|
|
|
subroutine gaussian_product_x(a,xa,b,xb,k,p,xp)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Gaussian product in 1D.
|
|
|
|
! e^{-a (x-x_A)^2} e^{-b (x-x_B)^2} = K_{ab}^x e^{-p (x-x_P)^2}
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
double precision , intent(in) :: a,b ! Exponents
|
|
|
|
double precision , intent(in) :: xa,xb ! Centers
|
|
|
|
double precision , intent(out) :: p ! New exponent
|
|
|
|
double precision , intent(out) :: xp ! New center
|
|
|
|
double precision , intent(out) :: k ! Constant
|
|
|
|
|
|
|
|
double precision :: p_inv
|
|
|
|
|
|
|
|
ASSERT (a>0.)
|
|
|
|
ASSERT (b>0.)
|
|
|
|
|
|
|
|
double precision :: xab, ab
|
|
|
|
|
|
|
|
p = a+b
|
|
|
|
p_inv = 1.d0/(a+b)
|
|
|
|
ab = a*b
|
|
|
|
xab = xa-xb
|
|
|
|
ab = ab*p_inv
|
|
|
|
k = ab*xab*xab
|
|
|
|
if (k > 40.d0) then
|
|
|
|
k=0.d0
|
|
|
|
return
|
|
|
|
endif
|
|
|
|
k = exp(-k)
|
|
|
|
xp = (a*xa+b*xb)*p_inv
|
|
|
|
end subroutine
|
|
|
|
|
|
|
|
|
2023-03-04 17:49:48 +01:00
|
|
|
!-
|
|
|
|
|
|
|
|
subroutine gaussian_product_x_v(a,xa,b,xb,k,p,xp,n_points)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Gaussian product in 1D with multiple xa
|
|
|
|
! e^{-a (x-x_A)^2} e^{-b (x-x_B)^2} = K_{ab}^x e^{-p (x-x_P)^2}
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
integer, intent(in) :: n_points
|
|
|
|
double precision , intent(in) :: a,b ! Exponents
|
|
|
|
double precision , intent(in) :: xa(n_points),xb ! Centers
|
|
|
|
double precision , intent(out) :: p(n_points) ! New exponent
|
|
|
|
double precision , intent(out) :: xp(n_points) ! New center
|
|
|
|
double precision , intent(out) :: k(n_points) ! Constant
|
|
|
|
|
|
|
|
double precision :: p_inv
|
|
|
|
integer :: ipoint
|
|
|
|
|
|
|
|
ASSERT (a>0.)
|
|
|
|
ASSERT (b>0.)
|
|
|
|
|
|
|
|
double precision :: xab, ab
|
|
|
|
|
|
|
|
p = a+b
|
|
|
|
p_inv = 1.d0/(a+b)
|
|
|
|
ab = a*b*p_inv
|
|
|
|
do ipoint = 1, n_points
|
|
|
|
xab = xa(ipoint)-xb
|
|
|
|
k(ipoint) = ab*xab*xab
|
|
|
|
if (k(ipoint) > 40.d0) then
|
|
|
|
k(ipoint)=0.d0
|
|
|
|
cycle
|
|
|
|
endif
|
|
|
|
k(ipoint) = exp(-k(ipoint))
|
|
|
|
xp(ipoint) = (a*xa(ipoint)+b*xb)*p_inv
|
|
|
|
enddo
|
|
|
|
end subroutine
|
|
|
|
|
|
|
|
|
2019-01-25 11:39:31 +01:00
|
|
|
|
2023-05-15 19:46:06 +02:00
|
|
|
subroutine multiply_poly_0c(b,c,nc,d,nd)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Multiply two polynomials
|
|
|
|
! D(t) += B(t)*C(t)
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
integer, intent(in) :: nc
|
|
|
|
integer, intent(out) :: nd
|
|
|
|
double precision, intent(in) :: b(0:0), c(0:nc)
|
|
|
|
double precision, intent(inout) :: d(0:0+nc)
|
|
|
|
|
|
|
|
integer :: ic
|
|
|
|
|
|
|
|
do ic = 0,nc
|
|
|
|
d(ic) = d(ic) + c(ic) * b(0)
|
|
|
|
enddo
|
|
|
|
|
|
|
|
do nd = nc,0,-1
|
|
|
|
if (d(nd) /= 0.d0) exit
|
|
|
|
enddo
|
|
|
|
|
|
|
|
end
|
|
|
|
|
|
|
|
subroutine multiply_poly_1c(b,c,nc,d,nd)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Multiply two polynomials
|
|
|
|
! D(t) += B(t)*C(t)
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
integer, intent(in) :: nc
|
|
|
|
integer, intent(out) :: nd
|
|
|
|
double precision, intent(in) :: b(0:1), c(0:nc)
|
|
|
|
double precision, intent(inout) :: d(0:1+nc)
|
|
|
|
|
|
|
|
integer :: ic, id
|
|
|
|
if(nc < 0) return
|
|
|
|
|
|
|
|
do ic = 0,nc
|
|
|
|
d( ic) = d( ic) + c(ic) * b(0)
|
|
|
|
d(1+ic) = d(1+ic) + c(ic) * b(1)
|
|
|
|
enddo
|
|
|
|
|
|
|
|
do nd = nc+1,0,-1
|
|
|
|
if (d(nd) /= 0.d0) exit
|
|
|
|
enddo
|
|
|
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
|
|
subroutine multiply_poly_2c(b,c,nc,d,nd)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Multiply two polynomials
|
|
|
|
! D(t) += B(t)*C(t)
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
integer, intent(in) :: nc
|
|
|
|
integer, intent(out) :: nd
|
|
|
|
double precision, intent(in) :: b(0:2), c(0:nc)
|
|
|
|
double precision, intent(inout) :: d(0:2+nc)
|
|
|
|
|
|
|
|
integer :: ic, id, k
|
|
|
|
if (nc <0) return
|
|
|
|
|
|
|
|
do ic = 0,nc
|
|
|
|
d( ic) = d( ic) + c(ic) * b(0)
|
|
|
|
d(1+ic) = d(1+ic) + c(ic) * b(1)
|
|
|
|
d(2+ic) = d(2+ic) + c(ic) * b(2)
|
|
|
|
enddo
|
|
|
|
|
|
|
|
do nd = nc+2,0,-1
|
|
|
|
if (d(nd) /= 0.d0) exit
|
|
|
|
enddo
|
|
|
|
|
|
|
|
end
|
|
|
|
|
|
|
|
subroutine multiply_poly_3c(b,c,nc,d,nd)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Multiply two polynomials
|
|
|
|
! D(t) += B(t)*C(t)
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
integer, intent(in) :: nc
|
|
|
|
integer, intent(out) :: nd
|
|
|
|
double precision, intent(in) :: b(0:3), c(0:nc)
|
|
|
|
double precision, intent(inout) :: d(0:3+nc)
|
|
|
|
|
|
|
|
integer :: ic, id
|
|
|
|
if (nc <0) return
|
|
|
|
|
|
|
|
do ic = 1,nc
|
|
|
|
d( ic) = d(1+ic) + c(ic) * b(0)
|
|
|
|
d(1+ic) = d(1+ic) + c(ic) * b(1)
|
|
|
|
d(2+ic) = d(1+ic) + c(ic) * b(2)
|
|
|
|
d(3+ic) = d(1+ic) + c(ic) * b(3)
|
|
|
|
enddo
|
|
|
|
|
|
|
|
do nd = nc+3,0,-1
|
|
|
|
if (d(nd) /= 0.d0) exit
|
|
|
|
enddo
|
|
|
|
|
|
|
|
end
|
|
|
|
|
2019-01-25 11:39:31 +01:00
|
|
|
|
|
|
|
|
|
|
|
subroutine multiply_poly(b,nb,c,nc,d,nd)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Multiply two polynomials
|
|
|
|
! D(t) += B(t)*C(t)
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
integer, intent(in) :: nb, nc
|
|
|
|
integer, intent(out) :: nd
|
|
|
|
double precision, intent(in) :: b(0:nb), c(0:nc)
|
|
|
|
double precision, intent(inout) :: d(0:nb+nc)
|
|
|
|
|
|
|
|
integer :: ndtmp
|
|
|
|
integer :: ib, ic, id, k
|
2023-05-15 19:46:06 +02:00
|
|
|
if(ior(nc,nb) < 0) return !False if nc>=0 and nb>=0
|
2019-01-25 11:39:31 +01:00
|
|
|
|
2023-05-15 19:46:06 +02:00
|
|
|
do ib=0,nb
|
|
|
|
do ic = 0,nc
|
2019-01-25 11:39:31 +01:00
|
|
|
d(ib+ic) = d(ib+ic) + c(ic) * b(ib)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
|
2023-05-15 19:46:06 +02:00
|
|
|
do nd = nb+nc,0,-1
|
|
|
|
if (d(nd) /= 0.d0) exit
|
2019-01-25 11:39:31 +01:00
|
|
|
enddo
|
|
|
|
|
|
|
|
end
|
|
|
|
|
2023-02-06 19:00:35 +01:00
|
|
|
subroutine multiply_poly_v(b,nb,c,nc,d,nd,n_points)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Multiply pairs of polynomials
|
|
|
|
! D(t) += B(t)*C(t)
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
integer, intent(in) :: nb, nc, n_points
|
|
|
|
integer, intent(in) :: nd
|
|
|
|
double precision, intent(in) :: b(n_points,0:nb), c(n_points,0:nc)
|
|
|
|
double precision, intent(inout) :: d(n_points,0:nd)
|
|
|
|
|
|
|
|
integer :: ib, ic, id, k, ipoint
|
|
|
|
if (nd < nb+nc) then
|
|
|
|
print *, nd, nb, nc
|
|
|
|
print *, irp_here, ': nd < nb+nc'
|
|
|
|
stop 1
|
|
|
|
endif
|
|
|
|
|
|
|
|
do ic = 0,nc
|
|
|
|
do ipoint=1, n_points
|
|
|
|
d(ipoint,ic) = d(ipoint,ic) + c(ipoint,ic) * b(ipoint,0)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
|
|
|
|
do ib=1,nb
|
|
|
|
do ipoint=1, n_points
|
|
|
|
d(ipoint, ib) = d(ipoint, ib) + c(ipoint,0) * b(ipoint, ib)
|
|
|
|
enddo
|
|
|
|
do ic = 1,nc
|
|
|
|
do ipoint=1, n_points
|
|
|
|
d(ipoint, ib+ic) = d(ipoint, ib+ic) + c(ipoint,ic) * b(ipoint, ib)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
enddo
|
2023-03-04 17:49:48 +01:00
|
|
|
|
2023-02-06 19:00:35 +01:00
|
|
|
end
|
|
|
|
|
2023-03-04 17:49:48 +01:00
|
|
|
|
2019-01-25 11:39:31 +01:00
|
|
|
subroutine add_poly(b,nb,c,nc,d,nd)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Add two polynomials
|
|
|
|
! D(t) += B(t)+C(t)
|
|
|
|
END_DOC
|
|
|
|
integer, intent(inout) :: nb, nc
|
|
|
|
integer, intent(out) :: nd
|
|
|
|
double precision, intent(in) :: b(0:nb), c(0:nc)
|
|
|
|
double precision, intent(out) :: d(0:nb+nc)
|
|
|
|
|
|
|
|
nd = nb+nc
|
|
|
|
integer :: ib, ic, id
|
|
|
|
do ib=0,max(nb,nc)
|
|
|
|
d(ib) = d(ib) + c(ib) + b(ib)
|
|
|
|
enddo
|
|
|
|
do while ( (d(nd) == 0.d0).and.(nd>=0) )
|
|
|
|
nd -= 1
|
|
|
|
if (nd < 0) then
|
|
|
|
exit
|
|
|
|
endif
|
|
|
|
enddo
|
|
|
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
subroutine add_poly_multiply(b,nb,cst,d,nd)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Add a polynomial multiplied by a constant
|
|
|
|
! D(t) += cst * B(t)
|
|
|
|
END_DOC
|
|
|
|
integer, intent(in) :: nb
|
|
|
|
integer, intent(inout) :: nd
|
|
|
|
double precision, intent(in) :: b(0:nb),cst
|
|
|
|
double precision, intent(inout) :: d(0:max(nb,nd))
|
|
|
|
|
|
|
|
nd = max(nd,nb)
|
|
|
|
if (nd /= -1) then
|
|
|
|
integer :: ib, ic, id
|
|
|
|
do ib=0,nb
|
|
|
|
d(ib) = d(ib) + cst*b(ib)
|
|
|
|
enddo
|
|
|
|
do while ( d(nd) == 0.d0 )
|
|
|
|
nd -= 1
|
|
|
|
if (nd < 0) then
|
|
|
|
exit
|
|
|
|
endif
|
|
|
|
enddo
|
|
|
|
endif
|
|
|
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
2023-02-06 19:00:35 +01:00
|
|
|
subroutine recentered_poly2_v(P_new, lda, x_A, LD_xA, x_P, a, P_new2, ldb, x_B, x_Q, b, n_points)
|
|
|
|
|
|
|
|
BEGIN_DOC
|
|
|
|
! Recenter two polynomials
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
implicit none
|
|
|
|
integer, intent(in) :: a(3), b(3), n_points, lda, ldb, LD_xA
|
|
|
|
double precision, intent(in) :: x_A(LD_xA,3), x_P(n_points,3), x_B(3), x_Q(n_points,3)
|
|
|
|
double precision, intent(out) :: P_new(n_points,0:lda,3),P_new2(n_points,0:ldb,3)
|
|
|
|
double precision :: binom_func
|
|
|
|
integer :: i,j,k,l, minab(3), maxab(3),ipoint, xyz
|
|
|
|
double precision, allocatable :: pows_a(:,:), pows_b(:,:)
|
|
|
|
double precision :: fa, fb
|
|
|
|
|
|
|
|
maxab(1:3) = max(a(1:3),b(1:3))
|
|
|
|
minab(1:3) = max(min(a(1:3),b(1:3)),(/0,0,0/))
|
|
|
|
|
|
|
|
allocate( pows_a(n_points,-2:maxval(maxab)+4), pows_b(n_points,-2:maxval(maxab)+4) )
|
|
|
|
|
|
|
|
do xyz=1,3
|
|
|
|
if ((a(xyz)<0).or.(b(xyz)<0) ) cycle
|
|
|
|
do ipoint=1,n_points
|
|
|
|
pows_a(ipoint,0) = 1.d0
|
|
|
|
pows_a(ipoint,1) = (x_P(ipoint,xyz) - x_A(ipoint,xyz))
|
|
|
|
pows_b(ipoint,0) = 1.d0
|
|
|
|
pows_b(ipoint,1) = (x_Q(ipoint,xyz) - x_B(xyz))
|
|
|
|
enddo
|
|
|
|
do i = 2,maxab(xyz)
|
|
|
|
do ipoint=1,n_points
|
|
|
|
pows_a(ipoint,i) = pows_a(ipoint,i-1)*pows_a(ipoint,1)
|
|
|
|
pows_b(ipoint,i) = pows_b(ipoint,i-1)*pows_b(ipoint,1)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
do ipoint=1,n_points
|
|
|
|
P_new (ipoint,0,xyz) = pows_a(ipoint,a(xyz))
|
|
|
|
P_new2(ipoint,0,xyz) = pows_b(ipoint,b(xyz))
|
|
|
|
enddo
|
|
|
|
do i = 1,min(minab(xyz),20)
|
|
|
|
fa = binom_transp(a(xyz)-i,a(xyz))
|
|
|
|
fb = binom_transp(b(xyz)-i,b(xyz))
|
|
|
|
do ipoint=1,n_points
|
|
|
|
P_new (ipoint,i,xyz) = fa * pows_a(ipoint,a(xyz)-i)
|
|
|
|
P_new2(ipoint,i,xyz) = fb * pows_b(ipoint,b(xyz)-i)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
do i = minab(xyz)+1,min(a(xyz),20)
|
|
|
|
fa = binom_transp(a(xyz)-i,a(xyz))
|
|
|
|
do ipoint=1,n_points
|
|
|
|
P_new (ipoint,i,xyz) = fa * pows_a(ipoint,a(xyz)-i)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
do i = minab(xyz)+1,min(b(xyz),20)
|
|
|
|
fb = binom_transp(b(xyz)-i,b(xyz))
|
|
|
|
do ipoint=1,n_points
|
|
|
|
P_new2(ipoint,i,xyz) = fb * pows_b(ipoint,b(xyz)-i)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
do i = 21,a(xyz)
|
|
|
|
fa = binom_func(a(xyz),a(xyz)-i)
|
|
|
|
do ipoint=1,n_points
|
|
|
|
P_new (ipoint,i,xyz) = fa * pows_a(ipoint,a(xyz)-i)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
do i = 21,b(xyz)
|
|
|
|
fb = binom_func(b(xyz),b(xyz)-i)
|
|
|
|
do ipoint=1,n_points
|
|
|
|
P_new2(ipoint,i,xyz) = fb * pows_b(ipoint,b(xyz)-i)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
|
|
|
|
end subroutine recentered_poly2_v
|
|
|
|
|
|
|
|
! ---
|
|
|
|
|
|
|
|
subroutine recentered_poly2_v0(P_new, lda, x_A, LD_xA, x_P, a, n_points)
|
|
|
|
|
|
|
|
BEGIN_DOC
|
|
|
|
!
|
|
|
|
! Recenter two polynomials. Special case for b=(0,0,0)
|
|
|
|
!
|
|
|
|
! (x - A)^a (x - B)^0 = (x - P + P - A)^a (x - Q + Q - B)^0
|
|
|
|
! = (x - P + P - A)^a
|
|
|
|
!
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
implicit none
|
|
|
|
integer, intent(in) :: a(3), n_points, lda, LD_xA
|
|
|
|
double precision, intent(in) :: x_A(LD_xA,3), x_P(n_points,3)
|
|
|
|
!double precision, intent(in) :: x_B(3), x_Q(n_points,3)
|
|
|
|
double precision, intent(out) :: P_new(n_points,0:lda,3)
|
|
|
|
!double precision, intent(out) :: P_new2(n_points,3)
|
|
|
|
|
|
|
|
integer :: i, j, k, l, xyz, ipoint, maxab(3)
|
|
|
|
double precision :: fa
|
|
|
|
double precision, allocatable :: pows_a(:,:)
|
|
|
|
!double precision, allocatable :: pows_b(:,:)
|
|
|
|
|
|
|
|
double precision :: binom_func
|
|
|
|
|
|
|
|
maxab(1:3) = max(a(1:3), (/0,0,0/))
|
|
|
|
|
|
|
|
allocate(pows_a(n_points,-2:maxval(maxab)+4))
|
|
|
|
!allocate(pows_b(n_points,-2:maxval(maxab)+4))
|
|
|
|
|
|
|
|
do xyz = 1, 3
|
|
|
|
if(a(xyz) < 0) cycle
|
|
|
|
|
|
|
|
do ipoint = 1, n_points
|
|
|
|
pows_a(ipoint,0) = 1.d0
|
|
|
|
pows_a(ipoint,1) = (x_P(ipoint,xyz) - x_A(ipoint,xyz))
|
|
|
|
!pows_b(ipoint,0) = 1.d0
|
|
|
|
!pows_b(ipoint,1) = (x_Q(ipoint,xyz) - x_B(xyz))
|
|
|
|
enddo
|
|
|
|
|
|
|
|
do i = 2, maxab(xyz)
|
|
|
|
do ipoint = 1, n_points
|
|
|
|
pows_a(ipoint,i) = pows_a(ipoint,i-1) * pows_a(ipoint,1)
|
|
|
|
!pows_b(ipoint,i) = pows_b(ipoint,i-1) * pows_b(ipoint,1)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
|
|
|
|
do ipoint = 1, n_points
|
|
|
|
P_new (ipoint,0,xyz) = pows_a(ipoint,a(xyz))
|
|
|
|
!P_new2(ipoint,xyz) = pows_b(ipoint,0)
|
|
|
|
enddo
|
|
|
|
do i = 1, min(a(xyz), 20)
|
|
|
|
fa = binom_transp(a(xyz)-i, a(xyz))
|
|
|
|
do ipoint = 1, n_points
|
|
|
|
P_new(ipoint,i,xyz) = fa * pows_a(ipoint,a(xyz)-i)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
do i = 21, a(xyz)
|
|
|
|
fa = binom_func(a(xyz), a(xyz)-i)
|
|
|
|
do ipoint = 1, n_points
|
|
|
|
P_new(ipoint,i,xyz) = fa * pows_a(ipoint,a(xyz)-i)
|
|
|
|
enddo
|
|
|
|
enddo
|
|
|
|
|
|
|
|
enddo !xyz
|
|
|
|
|
|
|
|
deallocate(pows_a)
|
|
|
|
!deallocate(pows_b)
|
|
|
|
|
|
|
|
end subroutine recentered_poly2_v0
|
2019-01-25 11:39:31 +01:00
|
|
|
|
|
|
|
subroutine recentered_poly2(P_new,x_A,x_P,a,P_new2,x_B,x_Q,b)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Recenter two polynomials
|
|
|
|
END_DOC
|
|
|
|
integer, intent(in) :: a,b
|
|
|
|
double precision, intent(in) :: x_A,x_P,x_B,x_Q
|
|
|
|
double precision, intent(out) :: P_new(0:a),P_new2(0:b)
|
|
|
|
double precision :: pows_a(-2:a+b+4), pows_b(-2:a+b+4)
|
|
|
|
double precision :: binom_func
|
|
|
|
integer :: i,j,k,l, minab, maxab
|
|
|
|
if ((a<0).or.(b<0) ) return
|
|
|
|
maxab = max(a,b)
|
|
|
|
minab = max(min(a,b),0)
|
|
|
|
pows_a(0) = 1.d0
|
|
|
|
pows_a(1) = (x_P - x_A)
|
|
|
|
pows_b(0) = 1.d0
|
|
|
|
pows_b(1) = (x_Q - x_B)
|
|
|
|
do i = 2,maxab
|
|
|
|
pows_a(i) = pows_a(i-1)*pows_a(1)
|
|
|
|
pows_b(i) = pows_b(i-1)*pows_b(1)
|
|
|
|
enddo
|
|
|
|
P_new (0) = pows_a(a)
|
|
|
|
P_new2(0) = pows_b(b)
|
|
|
|
do i = 1,min(minab,20)
|
|
|
|
P_new (i) = binom_transp(a-i,a) * pows_a(a-i)
|
|
|
|
P_new2(i) = binom_transp(b-i,b) * pows_b(b-i)
|
|
|
|
enddo
|
|
|
|
do i = minab+1,min(a,20)
|
|
|
|
P_new (i) = binom_transp(a-i,a) * pows_a(a-i)
|
|
|
|
enddo
|
|
|
|
do i = minab+1,min(b,20)
|
|
|
|
P_new2(i) = binom_transp(b-i,b) * pows_b(b-i)
|
|
|
|
enddo
|
|
|
|
do i = 101,a
|
|
|
|
P_new(i) = binom_func(a,a-i) * pows_a(a-i)
|
|
|
|
enddo
|
|
|
|
do i = 101,b
|
|
|
|
P_new2(i) = binom_func(b,b-i) * pows_b(b-i)
|
|
|
|
enddo
|
|
|
|
end
|
|
|
|
|
2023-02-06 19:00:35 +01:00
|
|
|
subroutine pol_modif_center(A_center, B_center, iorder, A_pol, B_pol)
|
|
|
|
|
|
|
|
BEGIN_DOC
|
|
|
|
!
|
|
|
|
! Transform the pol centerd on A:
|
|
|
|
! [ \sum_i ax_i (x-x_A)^i ] [ \sum_j ay_j (y-y_A)^j ] [ \sum_k az_k (z-z_A)^k ]
|
|
|
|
! to a pol centered on B
|
|
|
|
! [ \sum_i bx_i (x-x_B)^i ] [ \sum_j by_j (y-y_B)^j ] [ \sum_k bz_k (z-z_B)^k ]
|
|
|
|
!
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
! useful for max_dim
|
|
|
|
include 'constants.include.F'
|
|
|
|
|
|
|
|
implicit none
|
|
|
|
|
|
|
|
integer, intent(in) :: iorder(3)
|
|
|
|
double precision, intent(in) :: A_center(3), B_center(3)
|
|
|
|
double precision, intent(in) :: A_pol(0:max_dim, 3)
|
|
|
|
double precision, intent(out) :: B_pol(0:max_dim, 3)
|
|
|
|
|
|
|
|
integer :: i, Lmax
|
|
|
|
|
|
|
|
do i = 1, 3
|
|
|
|
Lmax = iorder(i)
|
|
|
|
call pol_modif_center_x( A_center(i), B_center(i), Lmax, A_pol(0:Lmax, i), B_pol(0:Lmax, i) )
|
|
|
|
enddo
|
|
|
|
|
|
|
|
return
|
|
|
|
end subroutine pol_modif_center
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
subroutine pol_modif_center_x(A_center, B_center, iorder, A_pol, B_pol)
|
|
|
|
|
|
|
|
BEGIN_DOC
|
|
|
|
!
|
|
|
|
! Transform the pol centerd on A:
|
|
|
|
! [ \sum_i ax_i (x-x_A)^i ]
|
|
|
|
! to a pol centered on B
|
|
|
|
! [ \sum_i bx_i (x-x_B)^i ]
|
|
|
|
!
|
|
|
|
! bx_i = \sum_{j=i}^{iorder} ax_j (x_B - x_A)^(j-i) j! / [ i! (j-i)! ]
|
|
|
|
! = \sum_{j=i}^{iorder} ax_j (x_B - x_A)^(j-i) binom_func(j,i)
|
|
|
|
!
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
implicit none
|
|
|
|
|
|
|
|
integer, intent(in) :: iorder
|
|
|
|
double precision, intent(in) :: A_center, B_center
|
|
|
|
double precision, intent(in) :: A_pol(0:iorder)
|
|
|
|
double precision, intent(out) :: B_pol(0:iorder)
|
|
|
|
|
|
|
|
integer :: i, j
|
|
|
|
double precision :: fact_tmp, dx
|
|
|
|
|
|
|
|
double precision :: binom_func
|
|
|
|
|
|
|
|
dx = B_center - A_center
|
|
|
|
|
|
|
|
do i = 0, iorder
|
|
|
|
fact_tmp = 0.d0
|
|
|
|
do j = i, iorder
|
|
|
|
fact_tmp += A_pol(j) * binom_func(j, i) * dx**dble(j-i)
|
|
|
|
enddo
|
|
|
|
B_pol(i) = fact_tmp
|
|
|
|
enddo
|
|
|
|
|
|
|
|
return
|
|
|
|
end subroutine pol_modif_center_x
|
|
|
|
|
|
|
|
|
2019-01-25 11:39:31 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
double precision function F_integral(n,p)
|
|
|
|
BEGIN_DOC
|
|
|
|
! function that calculates the following integral
|
|
|
|
! \int_{\-infty}^{+\infty} x^n \exp(-p x^2) dx
|
|
|
|
END_DOC
|
|
|
|
implicit none
|
|
|
|
integer :: n
|
|
|
|
double precision :: p
|
|
|
|
integer :: i,j
|
|
|
|
double precision :: accu,sqrt_p,fact_ratio,tmp,fact
|
|
|
|
include 'constants.include.F'
|
|
|
|
if(n < 0)then
|
|
|
|
F_integral = 0.d0
|
|
|
|
endif
|
|
|
|
if(iand(n,1).ne.0)then
|
|
|
|
F_integral = 0.d0
|
|
|
|
return
|
|
|
|
endif
|
|
|
|
sqrt_p = 1.d0/dsqrt(p)
|
|
|
|
if(n==0)then
|
|
|
|
F_integral = sqpi * sqrt_p
|
|
|
|
return
|
|
|
|
endif
|
|
|
|
F_integral = sqpi * 0.5d0**n * sqrt_p**(n+1) * fact(n)/fact(shiftr(n,1))
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
double precision function rint(n,rho)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
!.. math::
|
|
|
|
!
|
|
|
|
! \int_0^1 dx \exp(-p x^2) x^n
|
|
|
|
!
|
|
|
|
END_DOC
|
|
|
|
include 'constants.include.F'
|
|
|
|
double precision :: rho,u,rint1,v,val0,rint_large_n,u_inv
|
|
|
|
integer :: n,k
|
|
|
|
double precision :: two_rho_inv
|
|
|
|
|
|
|
|
if(n.eq.0)then
|
|
|
|
if(rho == 0.d0)then
|
|
|
|
rint=1.d0
|
|
|
|
else
|
|
|
|
u_inv=1.d0/dsqrt(rho)
|
|
|
|
u=rho*u_inv
|
2021-03-21 21:06:41 +01:00
|
|
|
rint=0.5d0*u_inv*sqpi*derf(u)
|
2019-01-25 11:39:31 +01:00
|
|
|
endif
|
|
|
|
return
|
|
|
|
endif
|
|
|
|
if(rho.lt.1.d0)then
|
|
|
|
rint=rint1(n,rho)
|
|
|
|
else
|
|
|
|
if(n.le.20)then
|
|
|
|
u_inv=1.d0/dsqrt(rho)
|
|
|
|
if(rho.gt.80.d0)then
|
|
|
|
v=0.d0
|
|
|
|
else
|
|
|
|
v=dexp(-rho)
|
|
|
|
endif
|
|
|
|
u=rho*u_inv
|
|
|
|
two_rho_inv = 0.5d0*u_inv*u_inv
|
2021-03-21 21:06:41 +01:00
|
|
|
val0=0.5d0*u_inv*sqpi*derf(u)
|
2019-01-25 11:39:31 +01:00
|
|
|
rint=(val0-v)*two_rho_inv
|
|
|
|
do k=2,n
|
|
|
|
rint=(rint*dfloat(k+k-1)-v)*two_rho_inv
|
|
|
|
enddo
|
|
|
|
else
|
|
|
|
rint=rint_large_n(n,rho)
|
|
|
|
endif
|
|
|
|
endif
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
double precision function rint_sum(n_pt_out,rho,d1)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Needed for the calculation of two-electron integrals.
|
|
|
|
END_DOC
|
|
|
|
include 'constants.include.F'
|
|
|
|
integer, intent(in) :: n_pt_out
|
|
|
|
double precision, intent(in) :: rho,d1(0:n_pt_out)
|
|
|
|
double precision :: u,rint1,v,val0,rint_large_n,u_inv
|
|
|
|
integer :: n,k,i
|
|
|
|
double precision :: two_rho_inv, rint_tmp, di
|
|
|
|
|
|
|
|
|
|
|
|
if(rho < 1.d0)then
|
|
|
|
|
|
|
|
if(rho == 0.d0)then
|
|
|
|
rint_sum=d1(0)
|
|
|
|
else
|
|
|
|
u_inv=1.d0/dsqrt(rho)
|
|
|
|
u=rho*u_inv
|
2021-03-21 21:06:41 +01:00
|
|
|
rint_sum=0.5d0*u_inv*sqpi*derf(u) *d1(0)
|
2019-01-25 11:39:31 +01:00
|
|
|
endif
|
|
|
|
|
|
|
|
do i=2,n_pt_out,2
|
|
|
|
n = shiftr(i,1)
|
|
|
|
rint_sum = rint_sum + d1(i)*rint1(n,rho)
|
|
|
|
enddo
|
|
|
|
|
|
|
|
else
|
|
|
|
|
|
|
|
if(rho.gt.80.d0)then
|
|
|
|
v=0.d0
|
|
|
|
else
|
|
|
|
v=dexp(-rho)
|
|
|
|
endif
|
|
|
|
|
|
|
|
u_inv=1.d0/dsqrt(rho)
|
|
|
|
u=rho*u_inv
|
|
|
|
two_rho_inv = 0.5d0*u_inv*u_inv
|
2021-03-21 21:06:41 +01:00
|
|
|
val0=0.5d0*u_inv*sqpi*derf(u)
|
2019-01-25 11:39:31 +01:00
|
|
|
rint_sum=val0*d1(0)
|
|
|
|
rint_tmp=(val0-v)*two_rho_inv
|
|
|
|
di = 3.d0
|
|
|
|
do i=2,min(n_pt_out,40),2
|
|
|
|
rint_sum = rint_sum + d1(i)*rint_tmp
|
|
|
|
rint_tmp = (rint_tmp*di-v)*two_rho_inv
|
|
|
|
di = di+2.d0
|
|
|
|
enddo
|
|
|
|
do i=42,n_pt_out,2
|
|
|
|
n = shiftr(i,1)
|
|
|
|
rint_sum = rint_sum + d1(i)*rint_large_n(n,rho)
|
|
|
|
enddo
|
|
|
|
|
|
|
|
endif
|
|
|
|
end
|
|
|
|
|
|
|
|
double precision function hermite(n,x)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Hermite polynomial
|
|
|
|
END_DOC
|
|
|
|
integer :: n,k
|
|
|
|
double precision :: h0,x,h1,h2
|
|
|
|
h0=1.d0
|
|
|
|
if(n.eq.0)then
|
|
|
|
hermite=h0
|
|
|
|
return
|
|
|
|
endif
|
|
|
|
h1=x+x
|
|
|
|
if(n.eq.1)then
|
|
|
|
hermite=h1
|
|
|
|
return
|
|
|
|
endif
|
|
|
|
do k=1,n-1
|
|
|
|
h2=(x+x)*h1-dfloat(k+k)*h0
|
|
|
|
h0=h1
|
|
|
|
h1=h2
|
|
|
|
enddo
|
|
|
|
hermite=h2
|
|
|
|
end
|
|
|
|
|
|
|
|
double precision function rint_large_n(n,rho)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Version of rint for large values of n
|
|
|
|
END_DOC
|
|
|
|
integer :: n,k,l
|
|
|
|
double precision :: rho,u,accu,eps,t1,t2,fact,alpha_k,rajout,hermite
|
|
|
|
u=dsqrt(rho)
|
|
|
|
accu=0.d0
|
|
|
|
k=0
|
|
|
|
eps=1.d0
|
|
|
|
do while (eps.gt.1.d-15)
|
|
|
|
t1=1.d0
|
|
|
|
do l=0,k
|
|
|
|
t1=t1*(n+n+l+1.d0)
|
|
|
|
enddo
|
|
|
|
t2=0.d0
|
|
|
|
do l=0,k
|
|
|
|
t2=t2+(-1.d0)**l/(fact(l+1)*fact(k-l))
|
|
|
|
enddo
|
|
|
|
alpha_k=t2*fact(k+1)*fact(k)*(-1.d0)**k
|
|
|
|
alpha_k= alpha_k/t1
|
|
|
|
rajout=(-1.d0)**k*u**k*hermite(k,u)*alpha_k/fact(k)
|
|
|
|
accu=accu+rajout
|
|
|
|
eps=dabs(rajout)/accu
|
|
|
|
k=k+1
|
|
|
|
enddo
|
|
|
|
rint_large_n=dexp(-rho)*accu
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
|
|
double precision function rint1(n,rho)
|
|
|
|
implicit none
|
|
|
|
BEGIN_DOC
|
|
|
|
! Standard version of rint
|
|
|
|
END_DOC
|
|
|
|
integer, intent(in) :: n
|
|
|
|
double precision, intent(in) :: rho
|
|
|
|
double precision, parameter :: eps=1.d-15
|
|
|
|
double precision :: rho_tmp, diff
|
|
|
|
integer :: k
|
|
|
|
rint1=inv_int(n+n+1)
|
|
|
|
rho_tmp = 1.d0
|
|
|
|
do k=1,20
|
|
|
|
rho_tmp = -rho_tmp*rho
|
|
|
|
diff=rho_tmp*fact_inv(k)*inv_int(shiftl(k+n,1)+1)
|
|
|
|
rint1=rint1+diff
|
|
|
|
if (dabs(diff) > eps) then
|
|
|
|
cycle
|
|
|
|
endif
|
|
|
|
return
|
|
|
|
enddo
|
|
|
|
write(*,*)'pb in rint1 k too large!'
|
|
|
|
stop 1
|
|
|
|
end
|
2023-03-04 17:49:48 +01:00
|
|
|
|
|
|
|
! ---
|
|
|
|
|
|
|
|
double precision function V_phi(n, m)
|
|
|
|
|
|
|
|
BEGIN_DOC
|
|
|
|
! Computes the angular $\phi$ part of the nuclear attraction integral:
|
|
|
|
!
|
|
|
|
! $\int_{0}^{2 \pi} \cos(\phi)^n \sin(\phi)^m d\phi$.
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
implicit none
|
|
|
|
integer, intent(in) :: n, m
|
|
|
|
|
|
|
|
integer :: i
|
|
|
|
double precision :: prod
|
|
|
|
|
|
|
|
double precision :: Wallis
|
|
|
|
|
|
|
|
prod = 1.d0
|
|
|
|
do i = 0, shiftr(n, 1)-1
|
|
|
|
prod = prod/ (1.d0 + dfloat(m+1)/dfloat(n-i-i-1))
|
|
|
|
enddo
|
|
|
|
V_phi = 4.d0 * prod * Wallis(m)
|
|
|
|
|
|
|
|
end function V_phi
|
|
|
|
|
|
|
|
! ---
|
|
|
|
|
|
|
|
double precision function V_theta(n, m)
|
|
|
|
|
|
|
|
BEGIN_DOC
|
|
|
|
! Computes the angular $\theta$ part of the nuclear attraction integral:
|
|
|
|
!
|
|
|
|
! $\int_{0}^{\pi} \cos(\theta)^n \sin(\theta)^m d\theta$
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
implicit none
|
|
|
|
include 'utils/constants.include.F'
|
|
|
|
integer, intent(in) :: n, m
|
|
|
|
|
|
|
|
integer :: i
|
|
|
|
double precision :: prod
|
|
|
|
|
|
|
|
double precision :: Wallis
|
|
|
|
|
|
|
|
V_theta = 0.d0
|
|
|
|
prod = 1.d0
|
|
|
|
do i = 0, shiftr(n, 1)-1
|
|
|
|
prod = prod / (1.d0 + dfloat(m+1)/dfloat(n-i-i-1))
|
|
|
|
enddo
|
|
|
|
V_theta = (prod + prod) * Wallis(m)
|
|
|
|
|
|
|
|
end function V_theta
|
|
|
|
|
|
|
|
! ---
|
|
|
|
|
|
|
|
double precision function Wallis(n)
|
|
|
|
|
|
|
|
BEGIN_DOC
|
|
|
|
! Wallis integral:
|
|
|
|
!
|
|
|
|
! $\int_{0}^{\pi} \cos(\theta)^n d\theta$.
|
|
|
|
END_DOC
|
|
|
|
|
|
|
|
implicit none
|
|
|
|
include 'utils/constants.include.F'
|
|
|
|
|
|
|
|
integer, intent(in) :: n
|
|
|
|
|
|
|
|
integer :: p
|
|
|
|
|
|
|
|
double precision :: fact
|
|
|
|
|
|
|
|
if(iand(n, 1) .eq. 0) then
|
|
|
|
|
|
|
|
Wallis = fact(shiftr(n, 1))
|
|
|
|
Wallis = pi * fact(n) / (dble(ibset(0_8, n)) * (Wallis + Wallis) * Wallis)
|
|
|
|
|
|
|
|
else
|
|
|
|
|
|
|
|
p = shiftr(n, 1)
|
|
|
|
Wallis = fact(p)
|
|
|
|
Wallis = dble(ibset(0_8, p+p)) * Wallis * Wallis / fact(p+p+1)
|
|
|
|
|
|
|
|
endif
|
|
|
|
|
|
|
|
end function Wallis
|
|
|
|
|
|
|
|
! ---
|
|
|
|
|