Latest working version of cacmb

development
Alex Selimov 3 years ago
parent a54215d024
commit 30dbd3bfc7

@ -2,8 +2,8 @@
.DEFAULT_GOAL := all
FC=mpif90
FFLAGS=-Wall -mcmodel=large -O0 -g -fbacktrace -fcheck=all -ffpe-trap=invalid,zero,overflow,underflow,denormal
FFLAGS=-O2 -mcmodel=large
#FFLAGS=-Wall -mcmodel=large -O0 -g -fbacktrace -fcheck=all -ffpe-trap=invalid,zero,overflow,underflow,denormal
FFLAGS=-O3 -g
OBJDIR=obj
SRCS := $(wildcard *.f90)

@ -33,7 +33,8 @@ module elements
integer :: atom_types = 0
!Variables for creating elements based on primitive cells
real(kind=dp) :: cubic_cell(3,8), fcc_cell(3,8), fcc_mat(3,3), fcc_inv(3,3), bcc_cell(3,8), bcc_mat(3,3), bcc_inv(3,3)
real(kind=dp) :: cubic_cell(3,8), fcc_cell(3,8), fcc_mat(3,3), fcc_inv(3,3), bcc_cell(3,8), bcc_mat(3,3), bcc_inv(3,3), &
cube20(3,20)
integer :: cubic_faces(4,6)
!Below are lattice type arrays which provide information on the general form of the elements.
@ -72,6 +73,29 @@ module elements
0.0_dp, 1.0_dp, 1.0_dp /), &
shape(fcc_cell))
!Now initialize the 20 node cube element
cube20 = reshape((/ 0.0_dp, 0.0_dp, 0.0_dp, &
1.0_dp, 0.0_dp, 0.0_dp, &
1.0_dp, 1.0_dp, 0.0_dp, &
0.0_dp, 1.0_dp, 0.0_dp, &
0.0_dp, 0.0_dp, 1.0_dp, &
1.0_dp, 0.0_dp, 1.0_dp, &
1.0_dp, 1.0_dp, 1.0_dp, &
0.0_dp, 1.0_dp, 1.0_dp, &
0.5_dp, 0.0_dp, 0.0_dp, &
1.0_dp, 0.5_dp, 0.0_dp, &
0.5_dp, 1.0_dp, 0.0_dp, &
0.0_dp, 0.5_dp, 0.0_dp, &
0.0_dp, 0.0_dp, 0.5_dp, &
1.0_dp, 0.0_dp, 0.5_dp, &
1.0_dp, 1.0_dp, 0.5_dp, &
0.0_dp, 1.0_dp, 0.5_dp, &
0.5_dp, 0.0_dp, 1.0_dp, &
1.0_dp, 0.5_dp, 1.0_dp, &
0.5_dp, 1.0_dp, 1.0_dp, &
0.0_dp, 0.5_dp, 1.0_dp /), &
shape(cube20))
!Now we create a list containing the list of vertices needed to describe the 6 cube faces
cubic_faces(:,1) = (/ 1, 2, 3, 4 /)
cubic_faces(:,2) = (/ 1, 2, 6, 5 /)
@ -384,6 +408,8 @@ module elements
ng_node(i) = 8
case('bcc')
ng_node(i) = 8
case('20fcc')
ng_node(i) = 20
end select
if(ng_node(i) > max_ng_node) max_ng_node = ng_node(i)
@ -462,6 +488,34 @@ module elements
end do
end do
end do
end do
end do
case('20fcc')
!Now loop over all the possible sites
do it = 1, esize
t = (1.0_dp*(it-1)-(esize-1)/2)/(1.0_dp*(esize-1)/2)
do is =1, esize
s = (1.0_dp*(is-1)-(esize-1)/2)/(1.0_dp*(esize-1)/2)
do ir = 1, esize
r = (1.0_dp*(ir-1) - (esize-1)/2)/(1.0_dp*(esize-1)/2)
call quad_rhombshape(r,s,t,a_shape)
do ibasis = 1, bnum
ia = ia + 1
do inod = 1, 20
type_interp(ia) = basis_type(ibasis,lat_type_temp)
r_interp(:,ia) = r_interp(:,ia) + a_shape(inod) * r_in(:,ibasis,inod)
if(present(data_interp)) then
!If data is present then interpolate data arrays as well
data_interp(1,ia) = data_interp(1,ia) + eng(ibasis, inod)*a_shape(inod)
data_interp(2:4,ia) = data_interp(2:4,ia) + f(:, ibasis, inod)*a_shape(inod)
data_interp(5:10,ia) = data_interp(5:10,ia) + v(:, ibasis, inod)*a_shape(inod)
end if
end do
end do
end do
end do
end do
@ -492,6 +546,42 @@ module elements
return
end subroutine rhombshape
subroutine quad_rhombshape(r,s,t, shape_fun)
!Shape function for 20 node quadratic rhombohedral elements
real(kind=8), intent(in) :: r, s, t
real(kind=8), intent(out) :: shape_fun(:)
!Corner nodes
shape_fun(1) = (1.0_dp-r)*(1.0_dp-s)*(1.0_dp-t)*(-r-s-t-2)/8.0_dp
shape_fun(2) = (1.0_dp+r)*(1.0_dp-s)*(1.0_dp-t)*(r-s-t-2)/8.0_dp
shape_fun(3) = (1.0_dp+r)*(1.0_dp+s)*(1.0_dp-t)*(r+s-t-2)/8.0_dp
shape_fun(4) = (1.0_dp-r)*(1.0_dp+s)*(1.0_dp-t)*(-r+s-t-2)/8.0_dp
shape_fun(5) = (1.0_dp-r)*(1.0_dp-s)*(1.0_dp+t)*(-r-s+t-2)/8.0_dp
shape_fun(6) = (1.0_dp+r)*(1.0_dp-s)*(1.0_dp+t)*(r-s+t-2)/8.0_dp
shape_fun(7) = (1.0_dp+r)*(1.0_dp+s)*(1.0_dp+t)*(r+s+t-2)/8.0_dp
shape_fun(8) = (1.0_dp-r)*(1.0_dp+s)*(1.0_dp+t)*(-r+s+t-2)/8.0_dp
!Side nodes, first node r is zero
shape_fun(9) = (1-r*r)*(1-s)*(1-t)/4.0_dp
shape_fun(11) = (1-r*r)*(1+s)*(1-t)/4.0_dp
shape_fun(17) = (1-r*r)*(1-s)*(1+t)/4.0_dp
shape_fun(19) = (1-r*r)*(1+s)*(1+t)/4.0_dp
!node s is zero
shape_fun(10) = (1+r)*(1-s*s)*(1-t)/4.0_dp
shape_fun(12) = (1-r)*(1-s*s)*(1-t)/4.0_dp
shape_fun(18) = (1+r)*(1-s*s)*(1+t)/4.0_dp
shape_fun(20) = (1-r)*(1-s*s)*(1+t)/4.0_dp
!node t is zero
shape_fun(13) = (1-r)*(1-s)*(1-t*t)/4.0_dp
shape_fun(14) = (1+r)*(1-s)*(1-t*t)/4.0_dp
shape_fun(15) = (1+r)*(1+s)*(1-t*t)/4.0_dp
shape_fun(16) = (1-r)*(1+s)*(1-t*t)/4.0_dp
end subroutine quad_rhombshape
subroutine delete_atoms(num, in_index)
!This subroutine deletes atoms from the atom arrays
integer, intent(in) :: num
@ -624,6 +714,9 @@ do i = 1, atom_num
else if(trim(adjustl(pos_string)) == '-inf') then
pos=box_bd(2*i-1)
else if(trim(adjustl(pos_string)) == 'mid') then
pos=(box_bd(2*i)+box_bd(2*i-1))/2
else if (trim(adjustl(pos_string)) == 'rand') then
call random_number(rand)
pos = (box_bd(2*i)-box_bd(2*i-1))*rand + box_bd(2*i-1)
@ -647,7 +740,6 @@ do i = 1, atom_num
end do
end if
read(cone, *) face
if ((face < 1).or.(face > 6)) stop "Current face number must be 1,2,3,4,5,6. Please check documentation"
!Now get the position
call offset_pos(ele, face, rand_ele_pos)
@ -670,6 +762,15 @@ do i = 1, atom_num
end if
pos = box_bd(2*i) - pos
else if ((index(pos_string,'-') > 0).and.(index(pos_string,'mid')>0)) then
!Now extract the number we are reducing from midinity
if(index(pos_string,'mid') < index(pos_string,'-')) then
read(pos_string(index(pos_string,'-')+1:), *, iostat=iospara) pos
else
read(pos_string(1:index(pos_string,'-')-1), *, iostat=iospara) pos
end if
pos = (box_bd(2*i)+box_bd(2*i-1))/2.0_dp - pos
else if ((index(pos_string,'+') > 0).and.(index(pos_string,'inf')>0)) then
!Now extract the number we are reducing from infinity
if(index(pos_string,'inf') < index(pos_string,'+')) then
@ -679,6 +780,15 @@ do i = 1, atom_num
end if
pos = box_bd(2*i-1) + pos
else if ((index(pos_string,'+') > 0).and.(index(pos_string,'mid')>0)) then
!Now extract the number we are reducing from midinity
if(index(pos_string,'mid') < index(pos_string,'+')) then
read(pos_string(index(pos_string,'+')+1:), *, iostat=iospara) pos
else
read(pos_string(1:index(pos_string,'+')-1), *, iostat=iospara) pos
end if
pos = (box_bd(2*i)+box_bd(2*i-1))/2.0_dp + pos
else if ((index(pos_string,'*') > 0).and.(index(pos_string,'inf')>0)) then
!Now extract the number we are reducing from infinity
if(index(pos_string,'inf') < index(pos_string,'*')) then
@ -809,6 +919,7 @@ do i = 1, atom_num
integer, intent(in) :: n,m !n-size of element arrays, m-size of atom arrays
integer :: allostat
print *, max_ng_node
!Allocate element arrays
if (n > 0) then
if (allocated(force_node)) then

@ -177,6 +177,7 @@ module io
do i = 1,ele_num
if(type_ele(i) == 'fcc') write_num = write_num + size_ele(i)**3
if(type_ele(i) == 'bcc') write_num = write_num + size_ele(i)**3
if(type_ele(i) == '20fcc') write_num = write_num + size_ele(i)**3
end do
!Write total number of atoms + elements
@ -212,7 +213,7 @@ module io
do i = 1, ele_num
call interpolate_atoms(type_ele(i), size_ele(i), lat_ele(i), r_node(:,:,:,i), type_interp, r_interp)
select case(trim(adjustl(type_ele(i))))
case('fcc','bcc')
case('fcc','bcc', '20fcc')
do iatom = 1, basisnum(lat_ele(i))*size_ele(i)**3
interp_num = interp_num+1
call apply_periodic(r_interp(:,iatom))
@ -255,6 +256,7 @@ module io
do i = 1,ele_num
if(type_ele(i) == 'fcc') write_num = write_num + size_ele(i)**3
if(type_ele(i) == 'bcc') write_num = write_num + size_ele(i)**3
if(type_ele(i) == '20fcc') write_num = write_num + size_ele(i)**3
end do
end if
!Write total number of atoms
@ -311,7 +313,7 @@ module io
call interpolate_atoms(type_ele(i), size_ele(i), lat_ele(i), r_node(:,:,:,i), type_interp, r_interp)
end if
select case(trim(adjustl(type_ele(i))))
case('fcc','bcc')
case('fcc','bcc', '20fcc')
if(write_dat) then
do iatom = 1, basisnum(lat_ele(i))*size_ele(i)**3
interp_num = interp_num+1
@ -562,7 +564,12 @@ module io
ip = 0
write(11,13)
do i = 1, ele_num
write(11, '(4i16)') i, basisnum(lat_ele(i)), 2, (size_ele(i)-1)
select case(type_ele(i))
case('fcc','bcc')
write(11, '(4i16)') i, basisnum(lat_ele(i)), 2, (size_ele(i)-1)
case('20fcc')
write(11, '(4i16)') i, basisnum(lat_ele(i)), 3, (size_ele(i)-1)
end select
do inod = 1, ng_node(lat_ele(i))
do ibasis = 1, basisnum(lat_ele(i))
ip = ip + 1
@ -624,7 +631,7 @@ module io
!Write out the elements, this is written in two stages, one line for the element and then 1 line for
!every basis at every node
do i = 1, ele_num
write(11, *) tag_ele(i), lat_ele(i), size_ele(i), sbox_ele(i), type_ele(i)
write(11, *) tag_ele(i), lat_ele(i), size_ele(i), sbox_ele(i), trim(adjustl(type_ele(i)))
do inod = 1, ng_node(lat_ele(i))
do ibasis =1, basisnum(lat_ele(i))
write(11,*) inod, ibasis, r_node(:, ibasis, inod, i)
@ -1003,7 +1010,7 @@ module io
integer :: i, j, in_eles, in_atoms, inbtypes(10), lat_type, ia, ie, inod, &
id, type_node, ilat, esize, tag, type, bnum, n, ibasis, ip, atom_type_map(100)
real(kind=dp) :: newdisplace(3), ra(3), in_lapa, ea, fa(3), va(6), &
ee(10,8), fe(3,10,8), ve(6,10,8), re(3,10,8), atomic_masses(10)
ee(10,20), fe(3,10,20), ve(6,10,20), re(3,10,20), atomic_masses(10)
character(len=100) :: textholder, fcc
character(len=1000) :: line
character(len=2) :: atomic_element
@ -1013,7 +1020,7 @@ module io
!Now initialize some important variables if they aren't defined
if (max_basisnum==0) max_basisnum = 10
if (max_ng_node==0) max_ng_node=8
if (max_ng_node==0) max_ng_node=20
fcc="fcc"
!Skip header comment lines
@ -1096,6 +1103,11 @@ module io
inbtypes(ibasis) = atom_type_map(inbtypes(ibasis))
end do
call lattice_map(bnum, inbtypes, n, 1.0_dp, lat_type)
if (n == 8)then
fcc="fcc"
else if (n == 20) then
fcc = "20fcc"
end if
call add_element(tag, fcc, esize+1, lat_type, sub_box_num, re)
call add_element_data(ele_num, ee, fe, ve)
end do
@ -1135,6 +1147,8 @@ module io
read(11,*) temp_box_bd(3:4), textholder, textholder
read(11,*) temp_box_bd(5:6), textholder, textholder
call grow_ele_arrays(0,atom_in)
print *, "Read in ", atom_in, " atoms from ", trim(adjustl(file))
!Shift the box boundaries if needed
do i = 1, 3

@ -67,7 +67,7 @@ module mode_create
end do
!Now get the rotated box vertex positions in lattice space. Should be integer units and get the new maxlen
select case(trim(adjustl(element_type)))
case('fcc')
case('fcc', '20fcc')
box_lat_vert = int(matmul(fcc_inv, matmul(orient_inv, box_vert)))+1
maxbd = maxval(matmul(orient,matmul(fcc_mat,box_lat_vert)),2)
case('bcc')
@ -93,7 +93,7 @@ module mode_create
end do
!Now get the rotated box vertex positions in lattice space. Should be integer units
select case(trim(adjustl(element_type)))
case('fcc')
case('fcc', '20fcc')
box_lat_vert = int(matmul(fcc_inv, matmul(orient_inv, box_vert)))+1
case('bcc')
box_lat_vert = int(matmul(bcc_inv, matmul(orient_inv, box_vert)))+1
@ -131,9 +131,11 @@ module mode_create
!Call the build function with the correct transformation matrix
select case(trim(adjustl(element_type)))
case('fcc')
call build_with_rhomb(box_lat_vert, fcc_mat)
call build_with_rhomb(box_lat_vert, fcc_mat, 8)
case('bcc')
call build_with_rhomb(box_lat_vert, bcc_mat)
call build_with_rhomb(box_lat_vert, bcc_mat, 8)
case('20fcc')
call build_with_rhomb(box_lat_vert, fcc_mat, 20)
case default
print *, "Element type ", trim(adjustl(element_type)), " not accepted in mode create, please specify a supported ",&
"element type"
@ -280,7 +282,7 @@ module mode_create
!Check special periodicity relations
select case(trim(adjustl(element_type)))
case('fcc')
case('fcc', '20fcc')
do i = 1,3
!Check if one of the directions is 110
if ((is_equal(abs(orient(i,1)), abs(orient(i,2))).and.(is_equal(orient(i,3),0.0_dp))).or.&
@ -331,19 +333,20 @@ module mode_create
!
end subroutine
subroutine build_with_rhomb(box_in_lat, transform_matrix)
subroutine build_with_rhomb(box_in_lat, transform_matrix, nn)
!This subroutine returns all the lattice points in the box in r_lat
!Inputs
integer, dimension(3,8), intent(in) :: box_in_lat !The box vertices transformed to lattice space
real(kind=dp), dimension(3,3), intent(in) :: transform_matrix !The transformation matrix from lattice_space to real space
integer, intent(in) :: nn
!Internal variables
integer :: i, inod, bd_in_lat(6), bd_in_array(6), ix, iy, iz, numlatpoints, ele(3,8), rzero(3), efill_size, &
vlat(3), temp_lat(3,8), m, n, o, j, k, nump_ele, efill_temp_lat(3,8), filzero(3), bd_ele_lat(6),&
efill_ele(3,8), ebd(6)
real(kind=dp) :: v(3), temp_nodes(3,1,8), r(3), centroid_bd(6)
integer :: i, inod, bd_in_lat(6), bd_in_array(6), ix, iy, iz, numlatpoints, ele(3,nn), rzero(3), efill_size, &
vlat(3), temp_lat(3,nn), m, n, o, j, k, nump_ele, efill_temp_lat(3,nn), filzero(3), &
bd_ele_lat(6), efill_ele(3,nn), ebd(6)
real(kind=dp) :: v(3), temp_nodes(3,1,nn), r(3), centroid_bd(6)
logical, allocatable :: lat_points(:,:,:)
logical :: node_in_bd(8), add, lat_points_ele(esize,esize,esize), intersect_bd(3)
logical :: node_in_bd(nn), add, lat_points_ele(esize,esize,esize), intersect_bd(3)
!Do some value initialization
max_esize = esize
@ -393,7 +396,12 @@ module mode_create
!If we are working with elements we have to use more complex code
!Initialize finite element
ele(:,:) = (esize-1) * cubic_cell(:,:)
select case(element_type)
case('fcc','bcc')
ele(:,:) = (esize-1) * cubic_cell(:,:)
case('20fcc')
ele(:,:) = (esize-1)*cube20
end select
!Make a 3 dimensional array of lattice points. This array is indexed by the integer lattice position.
!A value of true means that the coordinate is a lattice point which is within the box_bd
@ -423,7 +431,7 @@ module mode_create
do iy = 1, bd_in_array(2)
do ix = 1, bd_in_array(1)
node_in_bd(:) = .false.
do inod = 1, 8
do inod = 1, nn
vlat = ele(:,inod) + (/ ix, iy, iz /)
!Check to see if the node resides at a position containing a lattice point within the box
@ -445,16 +453,22 @@ module mode_create
!Now build the finite element region
lat_ele_num = 0
lat_atom_num = 0
allocate(r_lat(3,8,numlatpoints/esize), elat(numlatpoints/esize))
allocate(r_lat(3,nn,numlatpoints/esize), elat(numlatpoints/esize))
ele(:,:) = (esize-1) * cubic_cell(:,:)
select case(element_type)
case('fcc','bcc')
ele(:,:) = (esize-1) * cubic_cell(:,:)
case('20fcc')
ele(:,:) = (esize-1)*cube20
end select
!Redefined the second 3 indices as the number of elements that fit in the bounds
do i = 1, 3
bd_in_array(3+i) = bd_in_array(i)/esize
end do
!Now start the element at rzero
do inod=1, 8
do inod=1, nn
ele(:,inod) = ele(:,inod) + rzero
end do
do iz = -bd_in_array(6), bd_in_array(6)
@ -463,7 +477,7 @@ module mode_create
node_in_bd(:) = .false.
temp_nodes(:,:,:) = 0.0_dp
temp_lat(:,:) = 0
do inod = 1, 8
do inod = 1, nn
vlat= ele(:,inod) + (/ ix*(esize), iy*(esize), iz*(esize) /)
!Transform point back to real space for easier checking
! v = matmul(orient, matmul(transform_matrix,v))
@ -537,7 +551,7 @@ module mode_create
ze: do k = 1, (esize-efill_size)
ye: do j = 1, (esize-efill_size)
xe: do i = 1, (esize-efill_size)
do inod = 1,8
do inod = 1,nn
vlat = efill_ele(:,inod) + (/ i, j, k /)
if (.not.lat_points_ele(vlat(1),vlat(2),vlat(3))) cycle xe
do o = 1,3

@ -193,7 +193,7 @@ module mode_metric
integer :: i, inod, ibasis
npoints=0
rout = -huge(1.0_dp)
if(atom_num > 0) then
do i = 1, atom_num
rout(:,tag_atom(i)) = r_atom(:,i)

@ -41,7 +41,6 @@ module neighbors
box_len(i) = box_bd(2*i) - box_bd(2*i-1)
cell_num(i) = int(box_len(i)/(rc_off/2))+1
end do
print *, box_len, cell_num
!Initialize/allocate variables
cell_lim = 10
@ -49,13 +48,15 @@ module neighbors
!Now place points within cell
num_in_cell = 0
do i = 1, numinlist
!Check to see if the current point is a filler point and if so just skip it
if(r_list(1,i) < -huge(1.0_dp)+1) cycle
if(r_list(1,i) < box_bd(1)) cycle
!c is the position of the cell that the point belongs to
if(i == 11651203) print *, r_list(:,i), (r_list(1,i) < box_bd(1))
do j = 1, 3
c(j) = int((r_list(j,i)-box_bd(2*j-1))/(rc_off/2)) + 1
c(j) = int((r_list(j,i)-box_bd(2*j-1))/(rc_off)) + 1
end do
!Place the index in the correct position, growing if necessary
@ -102,7 +103,7 @@ module neighbors
pointloop: do i = 1, n
!First check to see if the point is a filler point, if so then skip it
if(r_list(1,i) < -Huge(-1.0_dp)+1) cycle
if(r_list(1,i) < box_bd(1)) cycle
!c is the position of the cell that the point
c = which_cell(:,i)

@ -644,7 +644,7 @@ module opt_group
if(group_ele_num > 0) then
orig_atom_num = atom_num
!Estimate number of atoms we are adding, this doesn't have to be exact
add_atom_num = group_ele_num*basisnum(lat_ele(element_index(1)))*size_ele(element_index(1))**3
add_atom_num = group_ele_num*max_basisnum*(max_esize)**3
call grow_ele_arrays(0,add_atom_num)
do i = 1, group_ele_num
ie = element_index(i)

Loading…
Cancel
Save