9
1
mirror of https://github.com/QuantumPackage/qp2.git synced 2024-06-17 10:05:18 +02:00

start working on NO

This commit is contained in:
Abdallah Ammar 2023-05-27 22:34:40 +02:00
parent 42c4d6ad56
commit b8bfab1d7c

View File

@ -1,3 +1,6 @@
! ---
BEGIN_PROVIDER [ double precision, normal_two_body_bi_orth, (mo_num, mo_num, mo_num, mo_num)]
BEGIN_DOC
@ -8,13 +11,16 @@ BEGIN_PROVIDER [ double precision, normal_two_body_bi_orth, (mo_num, mo_num, mo_
implicit none
integer :: i,h1,p1,h2,p2
integer :: hh1,hh2,pp1,pp2
integer :: i, h1, p1, h2, p2
integer :: hh1, hh2, pp1, pp2
integer :: Ne(2)
double precision :: hthree_aba, hthree_aaa, hthree_aab
double precision :: wall0, wall1
integer, allocatable :: occ(:,:)
integer(bit_kind), allocatable :: key_i_core(:,:)
double precision :: hthree_aba,hthree_aaa,hthree_aab
double precision :: wall0,wall1
print*,' Providing normal_two_body_bi_orth ...'
call wall_time(wall0)
PROVIDE N_int
@ -23,23 +29,21 @@ BEGIN_PROVIDER [ double precision, normal_two_body_bi_orth, (mo_num, mo_num, mo_
if(core_tc_op) then
do i = 1, N_int
key_i_core(i,1) = xor(ref_bitmask(i,1),core_bitmask(i,1))
key_i_core(i,2) = xor(ref_bitmask(i,2),core_bitmask(i,2))
key_i_core(i,1) = xor(ref_bitmask(i,1), core_bitmask(i,1))
key_i_core(i,2) = xor(ref_bitmask(i,2), core_bitmask(i,2))
enddo
call bitstring_to_list_ab(key_i_core,occ,Ne,N_int)
call bitstring_to_list_ab(key_i_core, occ, Ne, N_int)
else
call bitstring_to_list_ab(ref_bitmask,occ,Ne,N_int)
call bitstring_to_list_ab(ref_bitmask, occ, Ne, N_int)
endif
normal_two_body_bi_orth = 0.d0
print*,'Providing normal_two_body_bi_orth ...'
call wall_time(wall0)
normal_two_body_bi_orth(1:mo_num,1:mo_num,1:mo_num,1:mo_num) = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (hh1, h1, hh2, h2, pp1, p1, pp2, p2, hthree_aba, hthree_aab, hthree_aaa) &
!$OMP SHARED (N_int, n_act_orb, list_act, Ne, occ, normal_two_body_bi_orth)
!$OMP DO SCHEDULE (static)
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (hh1, h1, hh2, h2, pp1, p1, pp2, p2, hthree_aba, hthree_aab, hthree_aaa) &
!$OMP SHARED (N_int, n_act_orb, list_act, Ne, occ, normal_two_body_bi_orth)
!$OMP DO SCHEDULE (static)
do hh1 = 1, n_act_orb
h1 = list_act(hh1)
do pp1 = 1, n_act_orb
@ -48,50 +52,57 @@ BEGIN_PROVIDER [ double precision, normal_two_body_bi_orth, (mo_num, mo_num, mo_
h2 = list_act(hh2)
do pp2 = 1, n_act_orb
p2 = list_act(pp2)
! all contributions from the 3-e terms to the double excitations
! s1:(h1-->p1), s2:(h2-->p2) from the HF reference determinant
! opposite spin double excitations : s1 /= s2
call give_aba_contraction(N_int, h1, h2, p1, p2, Ne, occ, hthree_aba)
! same spin double excitations : s1 == s2
if(h1<h2.and.p1.gt.p2)then
! with opposite spin contributions
call give_aab_contraction(N_int, h2, h1, p1, p2, Ne, occ, hthree_aab) ! exchange h1<->h2
! same spin double excitations with same spin contributions
if(Ne(2).ge.3)then
call give_aaa_contraction(N_int, h2, h1, p1, p2, Ne, occ, hthree_aaa) ! exchange h1<->h2
else
hthree_aaa = 0.d0
endif
else
! with opposite spin contributions
call give_aab_contraction(N_int, h1, h2, p1, p2, Ne, occ, hthree_aab)
if(Ne(2).ge.3)then
if((h1 .lt. h2) .and. (p1 .gt. p2)) then
! with opposite spin contributions
call give_aab_contraction(N_int, h2, h1, p1, p2, Ne, occ, hthree_aab) ! exchange h1<->h2
! same spin double excitations with same spin contributions
call give_aaa_contraction(N_int, h1, h2, p1, p2, Ne, occ, hthree_aaa)
else
hthree_aaa = 0.d0
endif
if(Ne(2) .ge. 3) then
call give_aaa_contraction(N_int, h2, h1, p1, p2, Ne, occ, hthree_aaa) ! exchange h1<->h2
else
hthree_aaa = 0.d0
endif
else
! with opposite spin contributions
call give_aab_contraction(N_int, h1, h2, p1, p2, Ne, occ, hthree_aab)
if(Ne(2) .ge. 3) then
! same spin double excitations with same spin contributions
call give_aaa_contraction(N_int, h1, h2, p1, p2, Ne, occ, hthree_aaa)
else
hthree_aaa = 0.d0
endif
endif
normal_two_body_bi_orth(p2,h2,p1,h1) = 0.5d0*(hthree_aba + hthree_aab + hthree_aaa)
normal_two_body_bi_orth(p2,h2,p1,h1) = 0.5d0 * (hthree_aba + hthree_aab + hthree_aaa)
enddo
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
call wall_time(wall1)
print*,'Wall time for normal_two_body_bi_orth ',wall1-wall0
!$OMP END DO
!$OMP END PARALLEL
deallocate( occ )
deallocate( key_i_core )
call wall_time(wall1)
print*,' Wall time for normal_two_body_bi_orth ', wall1-wall0
END_PROVIDER
! ---
subroutine give_aba_contraction(Nint, h1, h2, p1, p2, Ne, occ, hthree)
@ -106,30 +117,41 @@ subroutine give_aba_contraction(Nint, h1, h2, p1, p2, Ne, occ, hthree)
!!!! double alpha/beta
hthree = 0.d0
do ii = 1, Ne(2) ! purely closed shell part
i = occ(ii,2)
call give_integrals_3_body_bi_ort(i ,p2,p1,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(i, p2, p1, i, h2, h1, integral)
int_direct = -1.d0 * integral
call give_integrals_3_body_bi_ort(p1,p2, i,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(p1, p2, i, i, h2, h1, integral)
int_exc_13 = -1.d0 * integral
call give_integrals_3_body_bi_ort(p2, i,p1,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(p2, i, p1, i, h2, h1, integral)
int_exc_12 = -1.d0 * integral
hthree += 2.d0 * int_direct - 1.d0 * ( int_exc_13 + int_exc_12)
hthree += 2.d0 * int_direct - 1.d0 * (int_exc_13 + int_exc_12)
enddo
do ii = Ne(2) + 1, Ne(1) ! purely open-shell part
i = occ(ii,1)
call give_integrals_3_body_bi_ort(i ,p2,p1,i,h2,h1,integral)
i = occ(ii,1)
call give_integrals_3_body_bi_ort(i, p2, p1, i, h2, h1, integral)
int_direct = -1.d0 * integral
call give_integrals_3_body_bi_ort(p1,p2, i,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(p1, p2, i, i, h2, h1, integral)
int_exc_13 = -1.d0 * integral
call give_integrals_3_body_bi_ort(p2, i,p1,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(p2, i, p1, i, h2, h1, integral)
int_exc_12 = -1.d0 * integral
hthree += 1.d0 * int_direct - 0.5d0* ( int_exc_13 + int_exc_12)
hthree += 1.d0 * int_direct - 0.5d0 * (int_exc_13 + int_exc_12)
enddo
end subroutine give_aba_contraction
return
end
! ---
BEGIN_PROVIDER [ double precision, normal_two_body_bi_orth_ab, (mo_num, mo_num, mo_num, mo_num)]
@ -152,29 +174,31 @@ BEGIN_PROVIDER [ double precision, normal_two_body_bi_orth_ab, (mo_num, mo_num,
allocate( key_i_core(N_int,2) )
allocate( occ(N_int*bit_kind_size,2) )
if(core_tc_op)then
do i = 1, N_int
key_i_core(i,1) = xor(ref_bitmask(i,1),core_bitmask(i,1))
key_i_core(i,2) = xor(ref_bitmask(i,2),core_bitmask(i,2))
enddo
call bitstring_to_list_ab(key_i_core,occ,Ne,N_int)
if(core_tc_op) then
do i = 1, N_int
key_i_core(i,1) = xor(ref_bitmask(i,1),core_bitmask(i,1))
key_i_core(i,2) = xor(ref_bitmask(i,2),core_bitmask(i,2))
enddo
call bitstring_to_list_ab(key_i_core,occ,Ne,N_int)
else
call bitstring_to_list_ab(ref_bitmask,occ,Ne,N_int)
call bitstring_to_list_ab(ref_bitmask,occ,Ne,N_int)
endif
normal_two_body_bi_orth_ab = 0.d0
do hh1 = 1, n_act_orb
h1 = list_act(hh1)
do pp1 = 1, n_act_orb
p1 = list_act(pp1)
do hh2 = 1, n_act_orb
h2 = list_act(hh2)
do pp2 = 1, n_act_orb
p2 = list_act(pp2)
call give_aba_contraction(N_int, h1, h2, p1, p2, Ne, occ, hthree)
normal_two_body_bi_orth_ab(p2,h2,p1,h1) = hthree
enddo
h1 = list_act(hh1)
do pp1 = 1, n_act_orb
p1 = list_act(pp1)
do hh2 = 1, n_act_orb
h2 = list_act(hh2)
do pp2 = 1, n_act_orb
p2 = list_act(pp2)
call give_aba_contraction(N_int, h1, h2, p1, p2, Ne, occ, hthree)
normal_two_body_bi_orth_ab(p2,h2,p1,h1) = hthree
enddo
enddo
enddo
enddo
enddo
deallocate( key_i_core )
@ -182,7 +206,7 @@ BEGIN_PROVIDER [ double precision, normal_two_body_bi_orth_ab, (mo_num, mo_num,
END_PROVIDER
! ---
BEGIN_PROVIDER [ double precision, normal_two_body_bi_orth_aa_bb, (n_act_orb, n_act_orb, n_act_orb, n_act_orb)]
@ -250,13 +274,14 @@ BEGIN_PROVIDER [ double precision, normal_two_body_bi_orth_aa_bb, (n_act_orb, n_
END_PROVIDER
! ---
subroutine give_aaa_contraction(Nint, h1, h2, p1, p2, Ne, occ, hthree)
BEGIN_DOC
! pure same spin contribution to same spin double excitation s1=h1,p1, s2=h2,p2, with s1==s2
! pure same spin contribution to same spin double excitation s1=h1,p1, s2=h2,p2, with s1==s2
END_DOC
use bitmasks ! you need to include the bitmasks_module.f90 features
implicit none
@ -270,48 +295,64 @@ subroutine give_aaa_contraction(Nint, h1, h2, p1, p2, Ne, occ, hthree)
hthree = 0.d0
do ii = 1, Ne(2) ! purely closed shell part
i = occ(ii,2)
call give_integrals_3_body_bi_ort(i ,p2,p1,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(i, p2, p1, i, h2, h1, integral)
int_direct = -1.d0 * integral
call give_integrals_3_body_bi_ort(p2,p1,i ,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(p2, p1, i, i, h2, h1, integral)
int_exc_l = -1.d0 * integral
call give_integrals_3_body_bi_ort(p1,i ,p2,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(p1, i, p2, i, h2, h1, integral)
int_exc_ll= -1.d0 * integral
call give_integrals_3_body_bi_ort(p2,i ,p1,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(p2, i, p1, i, h2, h1, integral)
int_exc_12= -1.d0 * integral
call give_integrals_3_body_bi_ort(p1,p2, i,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(p1, p2, i, i, h2, h1, integral)
int_exc_13= -1.d0 * integral
call give_integrals_3_body_bi_ort(i ,p1,p2,i,h2,h1,integral)
call give_integrals_3_body_bi_ort(i, p1, p2, i, h2, h1, integral)
int_exc_23= -1.d0 * integral
hthree += 1.d0 * int_direct + int_exc_l + int_exc_ll -( int_exc_12+ int_exc_13+ int_exc_23 )
hthree += 1.d0 * int_direct + int_exc_l + int_exc_ll - (int_exc_12 + int_exc_13 + int_exc_23)
enddo
do ii = Ne(2)+1,Ne(1) ! purely open-shell part
i = occ(ii,1)
call give_integrals_3_body_bi_ort(i ,p2,p1,i,h2,h1,integral)
int_direct = -1.d0 * integral
call give_integrals_3_body_bi_ort(p2,p1,i ,i,h2,h1,integral)
int_exc_l = -1.d0 * integral
call give_integrals_3_body_bi_ort(p1,i ,p2,i,h2,h1,integral)
int_exc_ll= -1.d0 * integral
call give_integrals_3_body_bi_ort(p2,i ,p1,i,h2,h1,integral)
int_exc_12= -1.d0 * integral
call give_integrals_3_body_bi_ort(p1,p2, i,i,h2,h1,integral)
int_exc_13= -1.d0 * integral
call give_integrals_3_body_bi_ort(i ,p1,p2,i,h2,h1,integral)
int_exc_23= -1.d0 * integral
hthree += 1.d0 * int_direct + 0.5d0 * (int_exc_l + int_exc_ll -( int_exc_12+ int_exc_13+ int_exc_23 ))
call give_integrals_3_body_bi_ort(i, p2, p1, i, h2, h1, integral)
int_direct = -1.d0 * integral
call give_integrals_3_body_bi_ort(p2, p1, i , i, h2, h1, integral)
int_exc_l = -1.d0 * integral
call give_integrals_3_body_bi_ort(p1, i, p2, i, h2, h1, integral)
int_exc_ll = -1.d0 * integral
call give_integrals_3_body_bi_ort(p2, i, p1, i, h2, h1, integral)
int_exc_12 = -1.d0 * integral
call give_integrals_3_body_bi_ort(p1, p2, i, i, h2, h1, integral)
int_exc_13 = -1.d0 * integral
call give_integrals_3_body_bi_ort(i, p1, p2, i, h2, h1, integral)
int_exc_23 = -1.d0 * integral
hthree += 1.d0 * int_direct + 0.5d0 * (int_exc_l + int_exc_ll - (int_exc_12 + int_exc_13 + int_exc_23))
enddo
end subroutine give_aaa_contraction
return
end
! ---
subroutine give_aab_contraction(Nint, h1, h2, p1, p2, Ne, occ, hthree)
implicit none
use bitmasks ! you need to include the bitmasks_module.f90 features
integer, intent(in) :: Nint, h1, h2, p1, p2
integer, intent(in) :: Ne(2), occ(Nint*bit_kind_size,2)
implicit none
integer, intent(in) :: Nint, h1, h2, p1, p2
integer, intent(in) :: Ne(2), occ(Nint*bit_kind_size,2)
double precision, intent(out) :: hthree
integer :: ii, i
double precision :: int_direct, int_exc_12, int_exc_13, int_exc_23
@ -320,11 +361,18 @@ subroutine give_aab_contraction(Nint, h1, h2, p1, p2, Ne, occ, hthree)
hthree = 0.d0
do ii = 1, Ne(2) ! purely closed shell part
i = occ(ii,2)
call give_integrals_3_body_bi_ort(p2,p1,i,h2,h1,i,integral)
call give_integrals_3_body_bi_ort(p2, p1, i, h2, h1, i, integral)
int_direct = -1.d0 * integral
call give_integrals_3_body_bi_ort(p1,p2,i,h2,h1,i,integral)
call give_integrals_3_body_bi_ort(p1, p2, i, h2, h1, i, integral)
int_exc_23= -1.d0 * integral
hthree += 1.d0 * int_direct - int_exc_23
hthree += 1.d0 * int_direct - int_exc_23
enddo
end subroutine give_aab_contraction
return
end
! ---