mirror of
https://github.com/QuantumPackage/qp2.git
synced 2024-12-26 21:33:30 +01:00
319 lines
10 KiB
Fortran
319 lines
10 KiB
Fortran
|
BEGIN_PROVIDER [integer , m_max_sm_7]
|
||
|
&BEGIN_PROVIDER [integer , n_max_sm_7]
|
||
|
&BEGIN_PROVIDER [integer , o_max_sm_7]
|
||
|
implicit none
|
||
|
BEGIN_DOC
|
||
|
! maximum value of the "m", "n" and "o" integer in the Jastrow function as in Eq. (4)
|
||
|
! of Schmidt,Moskowitz, JCP, 93, 4172 (1990) for the SM_7 version of Table IV
|
||
|
END_DOC
|
||
|
m_max_sm_7 = 4
|
||
|
n_max_sm_7 = 0
|
||
|
o_max_sm_7 = 4
|
||
|
END_PROVIDER
|
||
|
|
||
|
BEGIN_PROVIDER [integer , m_max_sm_9]
|
||
|
&BEGIN_PROVIDER [integer , n_max_sm_9]
|
||
|
&BEGIN_PROVIDER [integer , o_max_sm_9]
|
||
|
implicit none
|
||
|
BEGIN_DOC
|
||
|
! maximum value of the "m", "n" and "o" integer in the Jastrow function as in Eq. (4)
|
||
|
! of Schmidt,Moskowitz, JCP, 93, 4172 (1990) for the SM_9 version of Table IV
|
||
|
END_DOC
|
||
|
m_max_sm_9 = 4
|
||
|
n_max_sm_9 = 2
|
||
|
o_max_sm_9 = 4
|
||
|
END_PROVIDER
|
||
|
|
||
|
|
||
|
BEGIN_PROVIDER [integer , m_max_sm_17]
|
||
|
&BEGIN_PROVIDER [integer , n_max_sm_17]
|
||
|
&BEGIN_PROVIDER [integer , o_max_sm_17]
|
||
|
implicit none
|
||
|
BEGIN_DOC
|
||
|
! maximum value of the "m", "n" and "o" integer in the Jastrow function as in Eq. (4)
|
||
|
! of Schmidt,Moskowitz, JCP, 93, 4172 (1990) for the SM_17 version of Table IV
|
||
|
END_DOC
|
||
|
m_max_sm_17 = 6
|
||
|
n_max_sm_17 = 2
|
||
|
o_max_sm_17 = 6
|
||
|
END_PROVIDER
|
||
|
|
||
|
|
||
|
BEGIN_PROVIDER [ double precision, c_mn_o_sm_7, (0:m_max_sm_7,0:n_max_sm_7,0:o_max_sm_7,2:10)]
|
||
|
implicit none
|
||
|
BEGIN_DOC
|
||
|
!
|
||
|
!c_mn_o_7(0:4,0:4,2:10) = coefficient for the SM_7 correlation factor as given is Table IV of
|
||
|
! Schmidt,Moskowitz, JCP, 93, 4172 (1990)
|
||
|
! the first index (0:4) is the "m" integer for the 1e part
|
||
|
! the second index(0:0) is the "n" integer for the 1e part WHICH IS ALWAYS SET TO 0 FOR SM_7
|
||
|
! the third index (0:4) is the "o" integer for the 2e part
|
||
|
! the fourth index (2:10) is the nuclear charge of the atom
|
||
|
END_DOC
|
||
|
c_mn_o_sm_7 = 0.d0
|
||
|
integer :: i
|
||
|
do i = 2, 10 ! loop over nuclear charge
|
||
|
c_mn_o_sm_7(0,0,1,i) = 0.5d0 ! all the linear terms are set to 1/2 to satisfy the anti-parallel spin condition
|
||
|
enddo
|
||
|
! He atom
|
||
|
! two electron terms
|
||
|
c_mn_o_sm_7(0,0,2,2) = 0.50516d0
|
||
|
c_mn_o_sm_7(0,0,3,2) = -0.19313d0
|
||
|
c_mn_o_sm_7(0,0,4,2) = 0.30276d0
|
||
|
! one-electron terms
|
||
|
c_mn_o_sm_7(2,0,0,2) = -0.16995d0
|
||
|
c_mn_o_sm_7(3,0,0,2) = -0.34505d0
|
||
|
c_mn_o_sm_7(4,0,0,2) = -0.54777d0
|
||
|
! Ne atom
|
||
|
! two electron terms
|
||
|
c_mn_o_sm_7(0,0,2,10) = -0.792d0
|
||
|
c_mn_o_sm_7(0,0,3,10) = 1.05232d0
|
||
|
c_mn_o_sm_7(0,0,4,10) = -0.65615d0
|
||
|
! one-electron terms
|
||
|
c_mn_o_sm_7(2,0,0,10) = -0.13312d0
|
||
|
c_mn_o_sm_7(3,0,0,10) = -0.00131d0
|
||
|
c_mn_o_sm_7(4,0,0,10) = 0.09083d0
|
||
|
|
||
|
END_PROVIDER
|
||
|
|
||
|
BEGIN_PROVIDER [ double precision, c_mn_o_sm_9, (0:m_max_sm_9,0:n_max_sm_9,0:o_max_sm_9,2:10)]
|
||
|
implicit none
|
||
|
BEGIN_DOC
|
||
|
!
|
||
|
!c_mn_o_9(0:4,0:4,2:10) = coefficient for the SM_9 correlation factor as given is Table IV of
|
||
|
! Schmidt,Moskowitz, JCP, 93, 4172 (1990)
|
||
|
! the first index (0:4) is the "m" integer for the 1e part
|
||
|
! the second index(0:0) is the "n" integer for the 1e part WHICH IS ALWAYS SET TO 0 FOR SM_9
|
||
|
! the third index (0:4) is the "o" integer for the 2e part
|
||
|
! the fourth index (2:10) is the nuclear charge of the atom
|
||
|
END_DOC
|
||
|
c_mn_o_sm_9 = 0.d0
|
||
|
integer :: i
|
||
|
do i = 2, 10 ! loop over nuclear charge
|
||
|
c_mn_o_sm_9(0,0,1,i) = 0.5d0 ! all the linear terms are set to 1/2 to satisfy the anti-parallel spin condition
|
||
|
enddo
|
||
|
! He atom
|
||
|
! two electron terms
|
||
|
c_mn_o_sm_9(0,0,2,2) = 0.50516d0
|
||
|
c_mn_o_sm_9(0,0,3,2) = -0.19313d0
|
||
|
c_mn_o_sm_9(0,0,4,2) = 0.30276d0
|
||
|
! one-electron terms
|
||
|
c_mn_o_sm_9(2,0,0,2) = -0.16995d0
|
||
|
c_mn_o_sm_9(3,0,0,2) = -0.34505d0
|
||
|
c_mn_o_sm_9(4,0,0,2) = -0.54777d0
|
||
|
! Ne atom
|
||
|
! two electron terms
|
||
|
c_mn_o_sm_9(0,0,2,10) = -0.792d0
|
||
|
c_mn_o_sm_9(0,0,3,10) = 1.05232d0
|
||
|
c_mn_o_sm_9(0,0,4,10) = -0.65615d0
|
||
|
! one-electron terms
|
||
|
c_mn_o_sm_9(2,0,0,10) = -0.13312d0
|
||
|
c_mn_o_sm_9(3,0,0,10) = -0.00131d0
|
||
|
c_mn_o_sm_9(4,0,0,10) = 0.09083d0
|
||
|
|
||
|
END_PROVIDER
|
||
|
|
||
|
BEGIN_PROVIDER [ double precision, c_mn_o_sm_17, (0:m_max_sm_17,0:n_max_sm_17,0:o_max_sm_17,2:10)]
|
||
|
implicit none
|
||
|
BEGIN_DOC
|
||
|
!
|
||
|
!c_mn_o_17(0:4,0:4,2:10) = coefficient for the SM_17 correlation factor as given is Table IV of
|
||
|
! Schmidt,Moskowitz, JCP, 93, 4172 (1990)
|
||
|
! the first index (0:4) is the "m" integer for the 1e part
|
||
|
! the second index(0:0) is the "n" integer for the 1e part WHICH IS ALWAYS SET TO 0 FOR SM_17
|
||
|
! the third index (0:4) is the "o" integer for the 2e part
|
||
|
! the fourth index (2:10) is the nuclear charge of the atom
|
||
|
END_DOC
|
||
|
c_mn_o_sm_17 = 0.d0
|
||
|
integer :: i
|
||
|
do i = 2, 10 ! loop over nuclear charge
|
||
|
c_mn_o_sm_17(0,0,1,i) = 0.5d0 ! all the linear terms are set to 1/2 to satisfy the anti-parallel spin condition
|
||
|
enddo
|
||
|
! He atom
|
||
|
! two electron terms
|
||
|
c_mn_o_sm_17(0,0,2,2) = 0.09239d0
|
||
|
c_mn_o_sm_17(0,0,3,2) = -0.38664d0
|
||
|
c_mn_o_sm_17(0,0,4,2) = 0.95764d0
|
||
|
! one-electron terms
|
||
|
c_mn_o_sm_17(2,0,0,2) = 0.23208d0
|
||
|
c_mn_o_sm_17(3,0,0,2) = -0.45032d0
|
||
|
c_mn_o_sm_17(4,0,0,2) = 0.82777d0
|
||
|
c_mn_o_sm_17(2,2,0,2) = -4.15388d0
|
||
|
! ee-n terms
|
||
|
c_mn_o_sm_17(2,0,2,2) = 0.80622d0
|
||
|
c_mn_o_sm_17(2,2,2,2) = 10.19704d0
|
||
|
c_mn_o_sm_17(4,0,2,2) = -4.96259d0
|
||
|
c_mn_o_sm_17(2,0,4,2) = -1.35647d0
|
||
|
c_mn_o_sm_17(4,2,2,2) = -5.90907d0
|
||
|
c_mn_o_sm_17(6,0,2,2) = 0.90343d0
|
||
|
c_mn_o_sm_17(4,0,4,2) = 5.50739d0
|
||
|
c_mn_o_sm_17(2,2,4,2) = -0.03154d0
|
||
|
c_mn_o_sm_17(2,0,6,2) = -1.1051860
|
||
|
|
||
|
|
||
|
! Ne atom
|
||
|
! two electron terms
|
||
|
c_mn_o_sm_17(0,0,2,10) = -0.80909d0
|
||
|
c_mn_o_sm_17(0,0,3,10) = -0.00219d0
|
||
|
c_mn_o_sm_17(0,0,4,10) = 0.59188d0
|
||
|
! one-electron terms
|
||
|
c_mn_o_sm_17(2,0,0,10) = -0.00567d0
|
||
|
c_mn_o_sm_17(3,0,0,10) = 0.14011d0
|
||
|
c_mn_o_sm_17(4,0,0,10) = -0.05671d0
|
||
|
c_mn_o_sm_17(2,2,0,10) = -3.33767d0
|
||
|
! ee-n terms
|
||
|
c_mn_o_sm_17(2,0,2,10) = 1.95067d0
|
||
|
c_mn_o_sm_17(2,2,2,10) = 6.83340d0
|
||
|
c_mn_o_sm_17(4,0,2,10) = -3.29231d0
|
||
|
c_mn_o_sm_17(2,0,4,10) = -2.44998d0
|
||
|
c_mn_o_sm_17(4,2,2,10) = -2.13029d0
|
||
|
c_mn_o_sm_17(6,0,2,10) = 2.25768d0
|
||
|
c_mn_o_sm_17(4,0,4,10) = 1.97951d0
|
||
|
c_mn_o_sm_17(2,2,4,10) = -2.0924160
|
||
|
c_mn_o_sm_17(2,0,6,10) = 0.35493d0
|
||
|
|
||
|
END_PROVIDER
|
||
|
|
||
|
BEGIN_PROVIDER [ double precision, b_I_sm_90,(2:10)]
|
||
|
&BEGIN_PROVIDER [ double precision, d_I_sm_90,(2:10)]
|
||
|
implicit none
|
||
|
BEGIN_DOC
|
||
|
! "b_I" and "d_I" parameters of Eqs. (4) and (5) of Schmidt,Moskowitz, JCP, 93, 4172 (1990)
|
||
|
END_DOC
|
||
|
b_I_sm_90 = 1.d0
|
||
|
d_I_sm_90 = 1.d0
|
||
|
|
||
|
END_PROVIDER
|
||
|
|
||
|
subroutine get_full_sm_90_jastrow(r1,r2,rI,sm_j,i_charge, j_1e,j_2e,j_een,j_tot)
|
||
|
implicit none
|
||
|
double precision, intent(in) :: r1(3),r2(3),rI(3)
|
||
|
integer, intent(in) :: sm_j, i_charge
|
||
|
double precision, intent(out):: j_1e,j_2e,j_een,j_tot
|
||
|
BEGIN_DOC
|
||
|
! Jastrow function as in Eq. (4) of Schmidt,Moskowitz, JCP, 93, 4172 (1990)
|
||
|
! the i_charge variable is the integer specifying the charge of the atom for the Jastrow
|
||
|
! the sm_j integer variable represents the "quality" of the jastrow : sm_j = 7, 9, 17
|
||
|
END_DOC
|
||
|
double precision :: r_inucl,r_jnucl,r_ij,b_I, d_I
|
||
|
b_I = b_I_sm_90(i_charge)
|
||
|
d_I = d_I_sm_90(i_charge)
|
||
|
call get_rescaled_variables_j_sm_90(r1,r2,rI,b_I,d_I,r_inucl,r_jnucl,r_ij)
|
||
|
call jastrow_func_sm_90(r_inucl,r_jnucl,r_ij,sm_j,i_charge, j_1e,j_2e,j_een,j_tot)
|
||
|
end
|
||
|
|
||
|
subroutine get_rescaled_variables_j_sm_90(r1,r2,rI,b_I,d_I,r_inucl,r_jnucl,r_ij)
|
||
|
implicit none
|
||
|
BEGIN_DOC
|
||
|
! rescaled variables of Eq. (5) and (6) of Schmidt,Moskowitz, JCP, 93, 4172 (1990)
|
||
|
! the "b_I" and "d_I" parameters are the same as in Eqs. (5) and (6)
|
||
|
END_DOC
|
||
|
double precision, intent(in) :: r1(3),r2(3),rI(3)
|
||
|
double precision, intent(in) :: b_I, d_I
|
||
|
double precision, intent(out):: r_inucl,r_jnucl,r_ij
|
||
|
double precision :: rin, rjn, rij
|
||
|
integer :: i
|
||
|
rin = 0.d0
|
||
|
rjn = 0.d0
|
||
|
rij = 0.d0
|
||
|
do i = 1,3
|
||
|
rin += (r1(i) - rI(i)) * (r1(i) - rI(i))
|
||
|
rjn += (r2(i) - rI(i)) * (r2(i) - rI(i))
|
||
|
rij += (r2(i) - r1(i)) * (r2(i) - r1(i))
|
||
|
enddo
|
||
|
rin = dsqrt(rin)
|
||
|
rjn = dsqrt(rjn)
|
||
|
rij = dsqrt(rij)
|
||
|
r_inucl = b_I * rin/(1.d0 + b_I * rin)
|
||
|
r_jnucl = b_I * rjn/(1.d0 + b_I * rjn)
|
||
|
r_ij = d_I * rij/(1.d0 + b_I * rij)
|
||
|
end
|
||
|
|
||
|
subroutine jastrow_func_sm_90(r_inucl,r_jnucl,r_ij,sm_j,i_charge, j_1e,j_2e,j_een,j_tot)
|
||
|
implicit none
|
||
|
BEGIN_DOC
|
||
|
! Jastrow function as in Eq. (4) of Schmidt,Moskowitz, JCP, 93, 4172 (1990)
|
||
|
! Here the r_inucl, r_jnucl are the rescaled variables as defined in Eq. (5) with "b_I"
|
||
|
! r_ij is the rescaled variable as defined in Eq. (6) with "d_I"
|
||
|
! the i_charge variable is the integer specifying the charge of the atom for the Jastrow
|
||
|
! the sm_j integer variable represents the "quality" of the jastrow : sm_j = 7, 9, 17
|
||
|
!
|
||
|
! it returns the j_1e : sum of terms with "o" = "n" = 0, "m" /= 0,
|
||
|
! j_2e : sum of terms with "m" = "n" = 0, "o" /= 0,
|
||
|
! j_een : sum of terms with "m" /=0, "n" /= 0, "o" /= 0,
|
||
|
! j_tot : the total sum
|
||
|
END_DOC
|
||
|
double precision, intent(in) :: r_inucl,r_jnucl,r_ij
|
||
|
integer, intent(in) :: sm_j,i_charge
|
||
|
double precision, intent(out):: j_1e,j_2e,j_een,j_tot
|
||
|
j_1e = 0.D0
|
||
|
j_2e = 0.D0
|
||
|
j_een = 0.D0
|
||
|
double precision :: delta_mn,jastrow_sm_90_atomic
|
||
|
integer :: m,n,o
|
||
|
BEGIN_TEMPLATE
|
||
|
! pure 2e part
|
||
|
n = 0
|
||
|
m = 0
|
||
|
if(sm_j == $X )then
|
||
|
do o = 1, o_max_sm_$X
|
||
|
if(dabs(c_mn_o_sm_$X(m,n,o,i_charge)).lt.1.d-10)cycle
|
||
|
j_2e += c_mn_o_sm_$X(m,n,o,i_charge) * jastrow_sm_90_atomic(m,n,o,i_charge,r_inucl,r_jnucl,r_ij)
|
||
|
enddo
|
||
|
! else
|
||
|
! print*,'sm_j = ',sm_j
|
||
|
! print*,'not implemented, stop'
|
||
|
! stop
|
||
|
endif
|
||
|
! pure one-e part
|
||
|
o = 0
|
||
|
if(sm_j == $X)then
|
||
|
do n = 2, n_max_sm_$X
|
||
|
do m = 2, m_max_sm_$X
|
||
|
j_1e += c_mn_o_sm_$X(m,n,o,i_charge) * jastrow_sm_90_atomic(m,n,o,i_charge,r_inucl,r_jnucl,r_ij)
|
||
|
enddo
|
||
|
enddo
|
||
|
! else
|
||
|
! print*,'sm_j = ',sm_j
|
||
|
! print*,'not implemented, stop'
|
||
|
! stop
|
||
|
endif
|
||
|
! e-e-n part
|
||
|
if(sm_j == $X)then
|
||
|
do o = 1, o_max_sm_$X
|
||
|
do m = 2, m_max_sm_$X
|
||
|
do n = 2, n_max_sm_$X
|
||
|
j_een += c_mn_o_sm_$X(m,n,o,i_charge) * jastrow_sm_90_atomic(m,n,o,i_charge,r_inucl,r_jnucl,r_ij)
|
||
|
enddo
|
||
|
enddo
|
||
|
enddo
|
||
|
else
|
||
|
! print*,'sm_j = ',sm_j
|
||
|
! print*,'not implemented, stop'
|
||
|
! stop
|
||
|
endif
|
||
|
j_tot = j_1e + j_2e + j_een
|
||
|
SUBST [ X]
|
||
|
7 ;;
|
||
|
9 ;;
|
||
|
17 ;;
|
||
|
END_TEMPLATE
|
||
|
end
|
||
|
|
||
|
double precision function jastrow_sm_90_atomic(m,n,o,i_charge,r_inucl,r_jnucl,r_ij)
|
||
|
implicit none
|
||
|
BEGIN_DOC
|
||
|
! contribution to the function of Eq. (4) of Schmidt,Moskowitz, JCP, 93, 4172 (1990)
|
||
|
! for a given m,n,o and atom
|
||
|
END_DOC
|
||
|
double precision, intent(in) :: r_inucl,r_jnucl,r_ij
|
||
|
integer , intent(in) :: m,n,o,i_charge
|
||
|
double precision :: delta_mn
|
||
|
if(m==n)then
|
||
|
delta_mn = 0.5d0
|
||
|
else
|
||
|
delta_mn = 1.D0
|
||
|
endif
|
||
|
jastrow_sm_90_atomic = delta_mn * (r_inucl**m * r_jnucl**n + r_jnucl**m * r_inucl**n)*r_ij**o
|
||
|
end
|