def make_kpt(kpt): # kpt = kptj - kpti
# search for all possible ki and kj that has ki-kj+kpt=0
kk_match = numpy.einsum('ijx->ij', abs(kk_table + kpt)) < 1e-9
kpti_idx, kptj_idx = numpy.where(kk_todo & kk_match)
nkptj = len(kptj_idx)
log.debug1('kpt = %s', kpt)
log.debug2('kpti_idx = %s', kpti_idx)
log.debug2('kptj_idx = %s', kptj_idx)
kk_todo[kpti_idx,kptj_idx] = False
if swap_2e and not is_zero(kpt):
kk_todo[kptj_idx,kpti_idx] = False
max_memory1 = max_memory * (nkptj+1)/(nkptj+5)
blksize = max(int(max_memory1*4e6/(nkptj+5)/16/nao**2), 16)
bufR = numpy.empty((blksize*nao**2))
bufI = numpy.empty((blksize*nao**2))
# Use DF object to mimic KRHF/KUHF object in function get_coulG
mydf.exxdiv = exxdiv
vkcoulG = mydf.weighted_coulG(kpt, True, mydf.gs)
kptjs = kpts[kptj_idx]
# <r|-G+k_rs|s> = conj(<s|G-k_rs|r>) = conj(<s|G+k_sr|r>)
for k, pqkR, pqkI, p0, p1 \
in mydf.ft_loop(mydf.gs, kpt, kptjs, max_memory=max_memory1):
ki = kpti_idx[k]
kj = kptj_idx[k]
coulG = numpy.sqrt(vkcoulG[p0:p1])
# case 1: k_pq = (pi|iq)
#:v4 = numpy.einsum('ijL,lkL->ijkl', pqk, pqk.conj())
#:vk += numpy.einsum('ijkl,jk->il', v4, dm)
pqkR *= coulG
pqkI *= coulG
pLqR = lib.transpose(pqkR.reshape(nao,nao,-1), axes=(0,2,1), out=bufR)
pLqI = lib.transpose(pqkI.reshape(nao,nao,-1), axes=(0,2,1), out=bufI)
iLkR = numpy.empty((nao*(p1-p0),nao))
iLkI = numpy.empty((nao*(p1-p0),nao))
for i in range(nset):
iLkR, iLkI = zdotNN(pLqR.reshape(-1,nao), pLqI.reshape(-1,nao),
dmsR[i,kj], dmsI[i,kj], 1, iLkR, iLkI)
zdotNC(iLkR.reshape(nao,-1), iLkI.reshape(nao,-1),
pLqR.reshape(nao,-1).T, pLqI.reshape(nao,-1).T,
1, vkR[i,ki], vkI[i,ki], 1)
# case 2: k_pq = (iq|pi)
#:v4 = numpy.einsum('iLj,lLk->ijkl', pqk, pqk.conj())
#:vk += numpy.einsum('ijkl,li->kj', v4, dm)
if swap_2e and not is_zero(kpt):
iLkR = iLkR.reshape(nao,-1)
iLkI = iLkI.reshape(nao,-1)
for i in range(nset):
iLkR, iLkI = zdotNN(dmsR[i,ki], dmsI[i,ki], pLqR.reshape(nao,-1),
pLqI.reshape(nao,-1), 1, iLkR, iLkI)
zdotCN(pLqR.reshape(-1,nao).T, pLqI.reshape(-1,nao).T,
iLkR.reshape(-1,nao), iLkI.reshape(-1,nao),
1, vkR[i,kj], vkI[i,kj], 1)
pqkR = pqkI = coulG = pLqR = pLqI = iLkR = iLkI = None
def make_kpt(kpt): # kpt = kptj - kpti
# search for all possible ki and kj that has ki-kj+kpt=0
kk_match = numpy.einsum('ijx->ij', abs(kk_table + kpt)) < 1e-9
kpti_idx, kptj_idx = numpy.where(kk_todo & kk_match)
nkptj = len(kptj_idx)
log.debug1('kpt = %s', kpt)
log.debug2('kpti_idx = %s', kpti_idx)
log.debug2('kptj_idx = %s', kptj_idx)
kk_todo[kpti_idx,kptj_idx] = False
if swap_2e and not is_zero(kpt):
kk_todo[kptj_idx,kpti_idx] = False
max_memory1 = max_memory * (nkptj+1)/(nkptj+5)
blksize = max(int(max_memory1*4e6/(nkptj+5)/16/nao**2), 16)
bufR = numpy.empty((blksize*nao**2))
bufI = numpy.empty((blksize*nao**2))
# Use DF object to mimic KRHF/KUHF object in function get_coulG
mydf.exxdiv = exxdiv
vkcoulG = mydf.weighted_coulG(kpt, True, mydf.gs)
kptjs = kpts[kptj_idx]
# <r|-G+k_rs|s> = conj(<s|G-k_rs|r>) = conj(<s|G+k_sr|r>)
for k, pqkR, pqkI, p0, p1 \
in mydf.ft_loop(mydf.gs, kpt, kptjs, max_memory=max_memory1):
ki = kpti_idx[k]
kj = kptj_idx[k]
coulG = numpy.sqrt(vkcoulG[p0:p1])
# case 1: k_pq = (pi|iq)
#:v4 = numpy.einsum('ijL,lkL->ijkl', pqk, pqk.conj())
#:vk += numpy.einsum('ijkl,jk->il', v4, dm)
pqkR *= coulG
pqkI *= coulG
pLqR = lib.transpose(pqkR.reshape(nao,nao,-1), axes=(0,2,1), out=bufR)
pLqI = lib.transpose(pqkI.reshape(nao,nao,-1), axes=(0,2,1), out=bufI)
iLkR = numpy.empty((nao*(p1-p0),nao))
iLkI = numpy.empty((nao*(p1-p0),nao))
for i in range(nset):
iLkR, iLkI = zdotNN(pLqR.reshape(-1,nao), pLqI.reshape(-1,nao),
dmsR[i,kj], dmsI[i,kj], 1, iLkR, iLkI)
zdotNC(iLkR.reshape(nao,-1), iLkI.reshape(nao,-1),
pLqR.reshape(nao,-1).T, pLqI.reshape(nao,-1).T,
1, vkR[i,ki], vkI[i,ki], 1)
# case 2: k_pq = (iq|pi)
#:v4 = numpy.einsum('iLj,lLk->ijkl', pqk, pqk.conj())
#:vk += numpy.einsum('ijkl,li->kj', v4, dm)
if swap_2e and not is_zero(kpt):
iLkR = iLkR.reshape(nao,-1)
iLkI = iLkI.reshape(nao,-1)
for i in range(nset):
iLkR, iLkI = zdotNN(dmsR[i,ki], dmsI[i,ki], pLqR.reshape(nao,-1),
pLqI.reshape(nao,-1), 1, iLkR, iLkI)
zdotCN(pLqR.reshape(-1,nao).T, pLqI.reshape(-1,nao).T,
iLkR.reshape(-1,nao), iLkI.reshape(-1,nao),
1, vkR[i,kj], vkI[i,kj], 1)
pqkR = pqkI = coulG = pLqR = pLqI = iLkR = iLkI = None
def _update_vk_(vk, aoaoks, dms, coulG, weight, kpti_idx, kptj_idx, swap_2e):
dmsR, dmsI = dms
vkR, vkI = vk
nG = len(coulG)
n_dm = vkR.shape[0]
nao = vkR.shape[-1]
bufR = numpy.empty((nG * nao**2))
bufI = numpy.empty((nG * nao**2))
buf1R = numpy.empty((nG * nao**2))
buf1I = numpy.empty((nG * nao**2))
for k, aoao in enumerate(aoaoks):
ki = kpti_idx[k]
kj = kptj_idx[k]
# case 1: k_pq = (pi|iq)
#:v4 = numpy.einsum('ijL,lkL->ijkl', pqk, pqk.conj())
#:vk += numpy.einsum('ijkl,jk->il', v4, dm)
pLqR = numpy.ndarray((nao, nG, nao), buffer=bufR)
pLqI = numpy.ndarray((nao, nG, nao), buffer=bufI)
pLqR[:] = aoao.real.reshape(nG, nao, nao).transpose(1, 0, 2)
pLqI[:] = aoao.imag.reshape(nG, nao, nao).transpose(1, 0, 2)
iLkR = numpy.ndarray((nao, nG, nao), buffer=buf1R)
iLkI = numpy.ndarray((nao, nG, nao), buffer=buf1I)
for i in range(n_dm):
zdotNN(pLqR.reshape(-1, nao), pLqI.reshape(-1, nao),
dmsR[i, kj], dmsI[i, kj], 1, iLkR.reshape(-1, nao),
iLkI.reshape(-1, nao))
iLkR *= coulG.reshape(1, nG, 1)
iLkI *= coulG.reshape(1, nG, 1)
zdotNC(iLkR.reshape(nao, -1), iLkI.reshape(nao, -1),
pLqR.reshape(nao, -1).T,
pLqI.reshape(nao, -1).T, weight, vkR[i, ki], vkI[i, ki], 1)
# case 2: k_pq = (iq|pi)
#:v4 = numpy.einsum('iLj,lLk->ijkl', pqk, pqk.conj())
#:vk += numpy.einsum('ijkl,li->kj', v4, dm)
# <r|-G+k_rs|s> = conj(<s|G-k_rs|r>) = conj(<s|G+k_sr|r>)
if swap_2e:
for i in range(n_dm):
zdotNN(dmsR[i, ki], dmsI[i, ki], pLqR.reshape(nao, -1),
pLqI.reshape(nao, -1), 1, iLkR.reshape(nao, -1),
iLkI.reshape(nao, -1))
iLkR *= coulG.reshape(1, nG, 1)
iLkI *= coulG.reshape(1, nG, 1)
zdotCN(
pLqR.reshape(-1, nao).T,
pLqI.reshape(-1, nao).T, iLkR.reshape(-1, nao),
iLkI.reshape(-1, nao), weight, vkR[i, kj], vkI[i, kj], 1)
def contract_k(pLqR, pLqI, pjqR, pjqI):
# K ~ 'iLj,lLk*,li->kj' + 'lLk*,iLj,li->kj'
#:Lpq = LpqR + LpqI*1j
#:j3c = j3cR + j3cI*1j
#:for i in range(nset):
#: tmp = numpy.dot(dms[i], j3c.reshape(nao,-1))
#: vk1 = numpy.dot(Lpq.reshape(-1,nao).conj().T, tmp.reshape(-1,nao))
#: tmp = numpy.dot(dms[i], Lpq.reshape(nao,-1))
#: vk1+= numpy.dot(j3c.reshape(-1,nao).conj().T, tmp.reshape(-1,nao))
#: vkR[i] += vk1.real
#: vkI[i] += vk1.imag
nrow = pLqR.shape[1]
tmpR = numpy.ndarray((nao,nrow*nao), buffer=buf3R)
if k_real:
for i in range(nset):
tmpR = lib.ddot(dmsR[i], pjqR.reshape(nao,-1), 1, tmpR)
vk1R = lib.ddot(pLqR.reshape(-1,nao).T, tmpR.reshape(-1,nao))
vkR[i] += vk1R
if hermi:
vkR[i] += vk1R.T
else:
tmpR = lib.ddot(dmsR[i], pLqR.reshape(nao,-1), 1, tmpR)
lib.ddot(pjqR.reshape(-1,nao).T, tmpR.reshape(-1,nao),
1, vkR[i], 1)
else:
tmpI = numpy.ndarray((nao,nrow*nao), buffer=buf3I)
for i in range(nset):
tmpR, tmpI = zdotNN(dmsR[i], dmsI[i], pjqR.reshape(nao,-1),
pjqI.reshape(nao,-1), 1, tmpR, tmpI, 0)
vk1R, vk1I = zdotCN(pLqR.reshape(-1,nao).T, pLqI.reshape(-1,nao).T,
tmpR.reshape(-1,nao), tmpI.reshape(-1,nao))
vkR[i] += vk1R
vkI[i] += vk1I
if hermi:
vkR[i] += vk1R.T
vkI[i] -= vk1I.T
else:
tmpR, tmpI = zdotNN(dmsR[i], dmsI[i], pLqR.reshape(nao,-1),
pLqI.reshape(nao,-1), 1, tmpR, tmpI, 0)
zdotCN(pjqR.reshape(-1,nao).T, pjqI.reshape(-1,nao).T,
tmpR.reshape(-1,nao), tmpI.reshape(-1,nao),
1, vkR[i], vkI[i], 1)
def make_kpt(uniq_kptji_id): # kpt = kptj - kpti
kpt = uniq_kpts[uniq_kptji_id]
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
kLR = kLRs[uniq_kptji_id]
kLI = kLIs[uniq_kptji_id]
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao * (nao + 1) // 2
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('cint1e_ovlp_sph',
hermi=1,
kpts=adapted_kptjs)
for k, ji in enumerate(adapted_ji_idx):
ovlp[k] = lib.pack_tril(ovlp[k])
else:
aosym = 's1'
nao_pair = nao**2
mem_now = lib.current_memory()[0]
log.debug2('memory = %s', mem_now)
max_memory = max(2000, mydf.max_memory - mem_now)
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(
max(int(max_memory * .6 * 1e6 / 16 / naux / (nkptj + 1)), 1),
nao_pair)
shranges = _guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(
16, int(max_memory * .2 * 1e6 / 16 / buflen / (nkptj + 1)))
else:
Gblksize = max(
16, int(max_memory * .4 * 1e6 / 16 / buflen / (nkptj + 1)))
Gblksize = min(Gblksize, ngs, 16384)
pqkRbuf = numpy.empty(buflen * Gblksize)
pqkIbuf = numpy.empty(buflen * Gblksize)
# buf for ft_aopair
buf = numpy.zeros((nkptj, buflen * Gblksize), dtype=numpy.complex128)
col1 = 0
for istep, sh_range in enumerate(shranges):
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', \
istep+1, len(shranges), *sh_range)
bstart, bend, ncol = sh_range
col0, col1 = col1, col1 + ncol
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = numpy.asarray(feri['j3c/%d' % idx][:, col0:col1])
if is_zero(kpt):
for i, c in enumerate(vbar):
if c != 0:
v[i] -= c * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
v = None
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell,
Gv[p0:p1],
shls_slice,
aosym,
b,
gxyz[p0:p1],
Gvbase,
kpt,
adapted_kptjs,
out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG, ncol),
dtype=numpy.complex128,
order='F',
buffer=buf[k])
pqkR = numpy.ndarray((ncol, nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol, nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
aoao[:] = 0
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k][naux:], 1)
if not (is_zero(kpt)
and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k][naux:],
1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k][naux:], 1)
else:
shls_slice = (bstart, bend, 0, cell.nbas)
ni = ncol // nao
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell,
Gv[p0:p1],
shls_slice,
aosym,
b,
gxyz[p0:p1],
Gvbase,
kpt,
adapted_kptjs,
out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG, ni, nao),
dtype=numpy.complex128,
order='F',
buffer=buf[k])
pqkR = numpy.ndarray((ni, nao, nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ni, nao, nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.transpose(1, 2, 0)
pqkI[:] = aoao.imag.transpose(1, 2, 0)
aoao[:] = 0
pqkR = pqkR.reshape(-1, nG)
pqkI = pqkI.reshape(-1, nG)
zdotCN(kLR[p0:p1].T, kLI[p0:p1].T, pqkR.T, pqkI.T, -1,
j3cR[k][naux:], j3cI[k][naux:], 1)
naux0 = nauxs[uniq_kptji_id]
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = fuse(j3cR[k])
else:
v = fuse(j3cR[k] + j3cI[k] * 1j)
if j2c[uniq_kptji_id][0] == 'CD':
v = scipy.linalg.solve_triangular(j2c[uniq_kptji_id][1],
v,
lower=True,
overwrite_b=True)
else:
v = lib.dot(j2c[uniq_kptji_id][1], v)
feri['j3c/%d' % ji][:naux0, col0:col1] = v
naux0 = nauxs[uniq_kptji_id]
for k, ji in enumerate(adapted_ji_idx):
v = feri['j3c/%d' % ji][:naux0]
del (feri['j3c/%d' % ji])
feri['j3c/%d' % ji] = v
def _make_j3c(mydf, cell, auxcell, kptij_lst):
t1 = (time.clock(), time.time())
log = logger.Logger(mydf.stdout, mydf.verbose)
max_memory = max(2000, mydf.max_memory - lib.current_memory()[0])
fused_cell, fuse = fuse_auxcell(mydf, auxcell)
outcore.aux_e2(cell,
fused_cell,
mydf._cderi,
'cint3c2e_sph',
kptij_lst=kptij_lst,
dataname='j3c',
max_memory=max_memory)
t1 = log.timer_debug1('3c2e', *t1)
nao = cell.nao_nr()
naux = auxcell.nao_nr()
gs = mydf.gs
Gv, Gvbase, kws = cell.get_Gv_weights(gs)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngs = gxyz.shape[0]
kptis = kptij_lst[:, 0]
kptjs = kptij_lst[:, 1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('cint2c2e_sph', hermi=1, kpts=uniq_kpts)
kLRs = []
kLIs = []
# An alternative method to evalute j2c. This method might have larger numerical error?
# chgcell = make_modchg_basis(auxcell, mydf.eta)
# for k, kpt in enumerate(uniq_kpts):
# aoaux = ft_ao.ft_ao(chgcell, Gv, None, b, gxyz, Gvbase, kpt).T
# coulG = numpy.sqrt(mydf.weighted_coulG(kpt, False, gs))
# LkR = aoaux.real * coulG
# LkI = aoaux.imag * coulG
# j2caux = numpy.zeros_like(j2c[k])
# j2caux[naux:,naux:] = j2c[k][naux:,naux:]
# if is_zero(kpt): # kpti == kptj
# j2caux[naux:,naux:] -= lib.ddot(LkR, LkR.T)
# j2caux[naux:,naux:] -= lib.ddot(LkI, LkI.T)
# j2c[k] = j2c[k][:naux,:naux] - fuse(fuse(j2caux.T).T)
# vbar = fuse(mydf.auxbar(fused_cell))
# s = (vbar != 0).astype(numpy.double)
# j2c[k] -= numpy.einsum('i,j->ij', vbar, s)
# j2c[k] -= numpy.einsum('i,j->ij', s, vbar)
# else:
# j2cR, j2cI = zdotCN(LkR, LkI, LkR.T, LkI.T)
# j2caux[naux:,naux:] -= j2cR + j2cI * 1j
# j2c[k] = j2c[k][:naux,:naux] - fuse(fuse(j2caux.T).T)
#
# try:
# j2c[k] = scipy.linalg.cholesky(j2c[k], lower=True)
# except scipy.linalg.LinAlgError as e:
# msg =('===================================\n'
# 'J-metric not positive definite.\n'
# 'It is likely that gs is not enough.\n'
# '===================================')
# log.error(msg)
# raise scipy.linalg.LinAlgError('\n'.join([e.message, msg]))
# kLR = LkR.T
# kLI = LkI.T
# if not kLR.flags.c_contiguous: kLR = lib.transpose(LkR)
# if not kLI.flags.c_contiguous: kLI = lib.transpose(LkI)
# kLR *= coulG.reshape(-1,1)
# kLI *= coulG.reshape(-1,1)
# kLRs.append(kLR)
# kLIs.append(kLI)
# aoaux = LkR = LkI = kLR = kLI = coulG = None
for k, kpt in enumerate(uniq_kpts):
aoaux = ft_ao.ft_ao(fused_cell, Gv, None, b, gxyz, Gvbase, kpt).T
coulG = numpy.sqrt(mydf.weighted_coulG(kpt, False, gs))
LkR = aoaux.real * coulG
LkI = aoaux.imag * coulG
if is_zero(kpt): # kpti == kptj
j2c[k][naux:] -= lib.ddot(LkR[naux:], LkR.T)
j2c[k][naux:] -= lib.ddot(LkI[naux:], LkI.T)
j2c[k][:naux, naux:] = j2c[k][naux:, :naux].T
else:
j2cR, j2cI = zdotCN(LkR[naux:], LkI[naux:], LkR.T, LkI.T)
j2c[k][naux:] -= j2cR + j2cI * 1j
j2c[k][:naux, naux:] = j2c[k][naux:, :naux].T.conj()
j2c[k] = fuse(fuse(j2c[k]).T).T
try:
j2c[k] = ('CD', scipy.linalg.cholesky(j2c[k], lower=True))
except scipy.linalg.LinAlgError as e:
#msg =('===================================\n'
# 'J-metric not positive definite.\n'
# 'It is likely that gs is not enough.\n'
# '===================================')
#log.error(msg)
#raise scipy.linalg.LinAlgError('\n'.join([e.message, msg]))
w, v = scipy.linalg.eigh(j2c[k])
log.debug2('metric linear dependency for kpt %s', k)
log.debug2('cond = %.4g, drop %d bfns', w[0] / w[-1],
numpy.count_nonzero(w < LINEAR_DEP_THR))
v = v[:, w > LINEAR_DEP_THR].T.conj()
v /= numpy.sqrt(w[w > LINEAR_DEP_THR]).reshape(-1, 1)
j2c[k] = ('eig', v)
kLR = LkR[naux:].T
kLI = LkI[naux:].T
if not kLR.flags.c_contiguous: kLR = lib.transpose(LkR[naux:])
if not kLI.flags.c_contiguous: kLI = lib.transpose(LkI[naux:])
kLR *= coulG.reshape(-1, 1)
kLI *= coulG.reshape(-1, 1)
kLRs.append(kLR)
kLIs.append(kLI)
aoaux = LkR = LkI = kLR = kLI = coulG = None
nauxs = [v[1].shape[0] for v in j2c]
feri = h5py.File(mydf._cderi)
def make_kpt(uniq_kptji_id): # kpt = kptj - kpti
kpt = uniq_kpts[uniq_kptji_id]
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
kLR = kLRs[uniq_kptji_id]
kLI = kLIs[uniq_kptji_id]
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao * (nao + 1) // 2
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('cint1e_ovlp_sph',
hermi=1,
kpts=adapted_kptjs)
for k, ji in enumerate(adapted_ji_idx):
ovlp[k] = lib.pack_tril(ovlp[k])
else:
aosym = 's1'
nao_pair = nao**2
mem_now = lib.current_memory()[0]
log.debug2('memory = %s', mem_now)
max_memory = max(2000, mydf.max_memory - mem_now)
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(
max(int(max_memory * .6 * 1e6 / 16 / naux / (nkptj + 1)), 1),
nao_pair)
shranges = _guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(
16, int(max_memory * .2 * 1e6 / 16 / buflen / (nkptj + 1)))
else:
Gblksize = max(
16, int(max_memory * .4 * 1e6 / 16 / buflen / (nkptj + 1)))
Gblksize = min(Gblksize, ngs, 16384)
pqkRbuf = numpy.empty(buflen * Gblksize)
pqkIbuf = numpy.empty(buflen * Gblksize)
# buf for ft_aopair
buf = numpy.zeros((nkptj, buflen * Gblksize), dtype=numpy.complex128)
col1 = 0
for istep, sh_range in enumerate(shranges):
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', \
istep+1, len(shranges), *sh_range)
bstart, bend, ncol = sh_range
col0, col1 = col1, col1 + ncol
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = numpy.asarray(feri['j3c/%d' % idx][:, col0:col1])
if is_zero(kpt):
for i, c in enumerate(vbar):
if c != 0:
v[i] -= c * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
v = None
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell,
Gv[p0:p1],
shls_slice,
aosym,
b,
gxyz[p0:p1],
Gvbase,
kpt,
adapted_kptjs,
out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG, ncol),
dtype=numpy.complex128,
order='F',
buffer=buf[k])
pqkR = numpy.ndarray((ncol, nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol, nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
aoao[:] = 0
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k][naux:], 1)
if not (is_zero(kpt)
and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k][naux:],
1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k][naux:], 1)
else:
shls_slice = (bstart, bend, 0, cell.nbas)
ni = ncol // nao
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell,
Gv[p0:p1],
shls_slice,
aosym,
b,
gxyz[p0:p1],
Gvbase,
kpt,
adapted_kptjs,
out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG, ni, nao),
dtype=numpy.complex128,
order='F',
buffer=buf[k])
pqkR = numpy.ndarray((ni, nao, nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ni, nao, nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.transpose(1, 2, 0)
pqkI[:] = aoao.imag.transpose(1, 2, 0)
aoao[:] = 0
pqkR = pqkR.reshape(-1, nG)
pqkI = pqkI.reshape(-1, nG)
zdotCN(kLR[p0:p1].T, kLI[p0:p1].T, pqkR.T, pqkI.T, -1,
j3cR[k][naux:], j3cI[k][naux:], 1)
naux0 = nauxs[uniq_kptji_id]
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = fuse(j3cR[k])
else:
v = fuse(j3cR[k] + j3cI[k] * 1j)
if j2c[uniq_kptji_id][0] == 'CD':
v = scipy.linalg.solve_triangular(j2c[uniq_kptji_id][1],
v,
lower=True,
overwrite_b=True)
else:
v = lib.dot(j2c[uniq_kptji_id][1], v)
feri['j3c/%d' % ji][:naux0, col0:col1] = v
naux0 = nauxs[uniq_kptji_id]
for k, ji in enumerate(adapted_ji_idx):
v = feri['j3c/%d' % ji][:naux0]
del (feri['j3c/%d' % ji])
feri['j3c/%d' % ji] = v
for k, kpt in enumerate(uniq_kpts):
make_kpt(k)
feri.close()
def _make_j3c(mydf, cell, auxcell, kptij_lst, cderi_file):
log = logger.Logger(mydf.stdout, mydf.verbose)
t1 = t0 = (time.clock(), time.time())
fused_cell, fuse = fuse_auxcell(mydf, mydf.auxcell)
ao_loc = cell.ao_loc_nr()
nao = ao_loc[-1]
naux = auxcell.nao_nr()
nkptij = len(kptij_lst)
mesh = mydf.mesh
Gv, Gvbase, kws = cell.get_Gv_weights(mesh)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngrids = gxyz.shape[0]
kptis = kptij_lst[:,0]
kptjs = kptij_lst[:,1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
log.debug('Num uniq kpts %d', len(uniq_kpts))
log.debug2('uniq_kpts %s', uniq_kpts)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('int2c2e', hermi=1, kpts=uniq_kpts)
j2ctags = []
t1 = log.timer_debug1('2c2e', *t1)
swapfile = tempfile.NamedTemporaryFile(dir=os.path.dirname(cderi_file))
fswap = lib.H5TmpFile(swapfile.name)
# Unlink swapfile to avoid trash
swapfile = None
for k, kpt in enumerate(uniq_kpts):
coulG = mydf.weighted_coulG(kpt, False, mesh)
j2c[k] = fuse(fuse(j2c[k]).T).T.copy()
j2c_k = numpy.zeros_like(j2c[k])
for p0, p1 in mydf.mpi_prange(0, ngrids):
aoaux = ft_ao.ft_ao(fused_cell, Gv[p0:p1], None, b, gxyz[p0:p1], Gvbase, kpt).T
aoaux = fuse(aoaux)
LkR = numpy.asarray(aoaux.real, order='C')
LkI = numpy.asarray(aoaux.imag, order='C')
aoaux = None
if is_zero(kpt): # kpti == kptj
j2cR = lib.dot(LkR*coulG[p0:p1], LkR.T)
j2c_k += lib.dot(LkI*coulG[p0:p1], LkI.T, 1, j2cR, 1)
else:
# aoaux ~ kpt_ij, aoaux.conj() ~ kpt_kl
j2cR, j2cI = zdotCN(LkR*coulG[p0:p1], LkI*coulG[p0:p1], LkR.T, LkI.T)
j2c_k += j2cR + j2cI * 1j
LkR = LkI = None
j2c[k] -= mpi.allreduce(j2c_k)
try:
fswap['j2c/%d'%k] = scipy.linalg.cholesky(j2c[k], lower=True)
j2ctags.append('CD')
except scipy.linalg.LinAlgError:
w, v = scipy.linalg.eigh(j2c[k])
log.debug2('metric linear dependency for kpt %s', k)
log.debug2('cond = %.4g, drop %d bfns',
w[0]/w[-1], numpy.count_nonzero(w<mydf.linear_dep_threshold))
v1 = v[:,w>mydf.linear_dep_threshold].T.conj()
v1 /= numpy.sqrt(w[w>mydf.linear_dep_threshold]).reshape(-1,1)
fswap['j2c/%d'%k] = v1
if cell.dimension == 2 and cell.low_dim_ft_type != 'inf_vacuum':
idx = numpy.where(w < -mydf.linear_dep_threshold)[0]
if len(idx) > 0:
fswap['j2c-/%d'%k] = (v[:,idx]/numpy.sqrt(-w[idx])).conj().T
w = v = v1 = v2 = None
j2ctags.append('eig')
aoaux = kLR = kLI = j2cR = j2cI = coulG = None
j2c = None
aosym_s2 = numpy.einsum('ix->i', abs(kptis-kptjs)) < 1e-9
j_only = numpy.all(aosym_s2)
if gamma_point(kptij_lst):
dtype = 'f8'
else:
dtype = 'c16'
t1 = log.timer_debug1('aoaux and int2c', *t1)
# Estimates the buffer size based on the last contraction in G-space.
# This contraction requires to hold nkptj copies of (naux,?) array
# simultaneously in memory.
mem_now = max(comm.allgather(lib.current_memory()[0]))
max_memory = max(2000, mydf.max_memory - mem_now)
nkptj_max = max((uniq_inverse==x).sum() for x in set(uniq_inverse))
buflen = max(int(min(max_memory*.5e6/16/naux/(nkptj_max+2)/nao,
nao/3/mpi.pool.size)), 1)
chunks = (buflen, nao)
j3c_jobs = mpi_df.grids2d_int3c_jobs(cell, auxcell, kptij_lst, chunks, j_only)
log.debug1('max_memory = %d MB (%d in use) chunks %s',
max_memory, mem_now, chunks)
log.debug2('j3c_jobs %s', j3c_jobs)
if j_only:
int3c = wrap_int3c(cell, fused_cell, 'int3c2e', 's2', 1, kptij_lst)
else:
int3c = wrap_int3c(cell, fused_cell, 'int3c2e', 's1', 1, kptij_lst)
idxb = numpy.tril_indices(nao)
idxb = (idxb[0] * nao + idxb[1]).astype('i')
aux_loc = fused_cell.ao_loc_nr(fused_cell.cart)
def gen_int3c(job_id, ish0, ish1):
dataname = 'j3c-chunks/%d' % job_id
i0 = ao_loc[ish0]
i1 = ao_loc[ish1]
dii = i1*(i1+1)//2 - i0*(i0+1)//2
dij = (i1 - i0) * nao
if j_only:
buflen = max(8, int(max_memory*1e6/16/(nkptij*dii+dii)))
else:
buflen = max(8, int(max_memory*1e6/16/(nkptij*dij+dij)))
auxranges = balance_segs(aux_loc[1:]-aux_loc[:-1], buflen)
buflen = max([x[2] for x in auxranges])
buf = numpy.empty(nkptij*dij*buflen, dtype=dtype)
buf1 = numpy.empty(dij*buflen, dtype=dtype)
naux = aux_loc[-1]
for kpt_id, kptij in enumerate(kptij_lst):
key = '%s/%d' % (dataname, kpt_id)
if aosym_s2[kpt_id]:
shape = (naux, dii)
else:
shape = (naux, dij)
if gamma_point(kptij):
fswap.create_dataset(key, shape, 'f8')
else:
fswap.create_dataset(key, shape, 'c16')
naux0 = 0
for istep, auxrange in enumerate(auxranges):
log.alldebug2("aux_e1 job_id %d step %d", job_id, istep)
sh0, sh1, nrow = auxrange
sub_slice = (ish0, ish1, 0, cell.nbas, sh0, sh1)
if j_only:
mat = numpy.ndarray((nkptij,dii,nrow), dtype=dtype, buffer=buf)
else:
mat = numpy.ndarray((nkptij,dij,nrow), dtype=dtype, buffer=buf)
mat = int3c(sub_slice, mat)
for k, kptij in enumerate(kptij_lst):
h5dat = fswap['%s/%d'%(dataname,k)]
v = lib.transpose(mat[k], out=buf1)
if not j_only and aosym_s2[k]:
idy = idxb[i0*(i0+1)//2:i1*(i1+1)//2] - i0 * nao
out = numpy.ndarray((nrow,dii), dtype=v.dtype, buffer=mat[k])
v = numpy.take(v, idy, axis=1, out=out)
if gamma_point(kptij):
h5dat[naux0:naux0+nrow] = v.real
else:
h5dat[naux0:naux0+nrow] = v
naux0 += nrow
def ft_fuse(job_id, uniq_kptji_id, sh0, sh1):
kpt = uniq_kpts[uniq_kptji_id] # kpt = kptj - kpti
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
Gaux = ft_ao.ft_ao(fused_cell, Gv, None, b, gxyz, Gvbase, kpt).T
Gaux = fuse(Gaux)
Gaux *= mydf.weighted_coulG(kpt, False, mesh)
kLR = lib.transpose(numpy.asarray(Gaux.real, order='C'))
kLI = lib.transpose(numpy.asarray(Gaux.imag, order='C'))
j2c = numpy.asarray(fswap['j2c/%d'%uniq_kptji_id])
j2ctag = j2ctags[uniq_kptji_id]
naux0 = j2c.shape[0]
if ('j2c-/%d' % uniq_kptji_id) in fswap:
j2c_negative = numpy.asarray(fswap['j2c-/%d'%uniq_kptji_id])
else:
j2c_negative = None
if is_zero(kpt):
aosym = 's2'
else:
aosym = 's1'
if aosym == 's2' and cell.dimension == 3:
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('int1e_ovlp', hermi=1, kpts=adapted_kptjs)
ovlp = [lib.pack_tril(s) for s in ovlp]
j3cR = [None] * nkptj
j3cI = [None] * nkptj
i0 = ao_loc[sh0]
i1 = ao_loc[sh1]
for k, idx in enumerate(adapted_ji_idx):
key = 'j3c-chunks/%d/%d' % (job_id, idx)
v = fuse(numpy.asarray(fswap[key]))
if aosym == 's2' and cell.dimension == 3:
for i in numpy.where(vbar != 0)[0]:
v[i] -= vbar[i] * ovlp[k][i0*(i0+1)//2:i1*(i1+1)//2].ravel()
j3cR[k] = numpy.asarray(v.real, order='C')
if v.dtype == numpy.complex128:
j3cI[k] = numpy.asarray(v.imag, order='C')
v = None
ncol = j3cR[0].shape[1]
Gblksize = max(16, int(max_memory*1e6/16/ncol/(nkptj+1))) # +1 for pqkRbuf/pqkIbuf
Gblksize = min(Gblksize, ngrids, 16384)
pqkRbuf = numpy.empty(ncol*Gblksize)
pqkIbuf = numpy.empty(ncol*Gblksize)
buf = numpy.empty(nkptj*ncol*Gblksize, dtype=numpy.complex128)
log.alldebug2(' blksize (%d,%d)', Gblksize, ncol)
if aosym == 's2':
shls_slice = (sh0, sh1, 0, sh1)
else:
shls_slice = (sh0, sh1, 0, cell.nbas)
for p0, p1 in lib.prange(0, ngrids, Gblksize):
dat = ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym, b,
gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = dat[k].reshape(nG,ncol)
pqkR = numpy.ndarray((ncol,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k], 1)
for k, idx in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = j3cR[k]
else:
v = j3cR[k] + j3cI[k] * 1j
if j2ctag == 'CD':
v = scipy.linalg.solve_triangular(j2c, v, lower=True, overwrite_b=True)
fswap['j3c-chunks/%d/%d'%(job_id,idx)][:naux0] = v
else:
fswap['j3c-chunks/%d/%d'%(job_id,idx)][:naux0] = lib.dot(j2c, v)
# low-dimension systems
if j2c_negative is not None:
fswap['j3c-/%d/%d'%(job_id,idx)] = lib.dot(j2c_negative, v)
mpi_df._assemble(mydf, kptij_lst, j3c_jobs, gen_int3c, ft_fuse, cderi_file, fswap, log)
def _make_j3c(mydf, cell, auxcell, kptij_lst, cderi_file):
t1 = (time.clock(), time.time())
log = logger.Logger(mydf.stdout, mydf.verbose)
max_memory = max(2000, mydf.max_memory-lib.current_memory()[0])
fused_cell, fuse = fuse_auxcell(mydf, auxcell)
# Create swap file to avoid huge cderi_file. see also function
# pyscf.pbc.df.df._make_j3c
swapfile = tempfile.NamedTemporaryFile(dir=os.path.dirname(cderi_file))
fswap = lib.H5TmpFile(swapfile.name)
# Unlink swapfile to avoid trash
swapfile = None
outcore._aux_e2(cell, fused_cell, fswap, 'int3c2e', aosym='s2',
kptij_lst=kptij_lst, dataname='j3c-junk', max_memory=max_memory)
t1 = log.timer_debug1('3c2e', *t1)
nao = cell.nao_nr()
naux = auxcell.nao_nr()
mesh = mydf.mesh
Gv, Gvbase, kws = cell.get_Gv_weights(mesh)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngrids = gxyz.shape[0]
kptis = kptij_lst[:,0]
kptjs = kptij_lst[:,1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
log.debug('Num uniq kpts %d', len(uniq_kpts))
log.debug2('uniq_kpts %s', uniq_kpts)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('int2c2e', hermi=1, kpts=uniq_kpts)
for k, kpt in enumerate(uniq_kpts):
aoaux = ft_ao.ft_ao(fused_cell, Gv, None, b, gxyz, Gvbase, kpt).T
aoaux = fuse(aoaux)
coulG = mydf.weighted_coulG(kpt, False, mesh)
LkR = numpy.asarray(aoaux.real, order='C')
LkI = numpy.asarray(aoaux.imag, order='C')
j2c_k = fuse(fuse(j2c[k]).T).T.copy()
if is_zero(kpt): # kpti == kptj
j2c_k -= lib.dot(LkR*coulG, LkR.T)
j2c_k -= lib.dot(LkI*coulG, LkI.T)
else:
# aoaux ~ kpt_ij, aoaux.conj() ~ kpt_kl
j2cR, j2cI = zdotCN(LkR*coulG, LkI*coulG, LkR.T, LkI.T)
j2c_k -= j2cR + j2cI * 1j
fswap['j2c/%d'%k] = j2c_k
aoaux = LkR = LkI = j2cR = j2cI = coulG = None
j2c = None
def cholesky_decomposed_metric(uniq_kptji_id):
j2c = numpy.asarray(fswap['j2c/%d'%uniq_kptji_id])
j2c_negative = None
# Note large difference may be found in results between the CD/eig treatments.
# In some systems, small integral errors can lead to different treatments of
# linear dependency which can be observed in the total energy/orbital energy
# around 4th decimal place.
# try:
# j2c = scipy.linalg.cholesky(j2c, lower=True)
# j2ctag = 'CD'
# except scipy.linalg.LinAlgError as e:
#
# Abandon CD treatment for better numerical stability
w, v = scipy.linalg.eigh(j2c)
log.debug('MDF metric for kpt %s cond = %.4g, drop %d bfns',
uniq_kptji_id, w[-1]/w[0], numpy.count_nonzero(w<mydf.linear_dep_threshold))
v1 = v[:,w>mydf.linear_dep_threshold].T.conj()
v1 /= numpy.sqrt(w[w>mydf.linear_dep_threshold]).reshape(-1,1)
j2c = v1
if cell.dimension == 2 and cell.low_dim_ft_type != 'inf_vacuum':
idx = numpy.where(w < -mydf.linear_dep_threshold)[0]
if len(idx) > 0:
j2c_negative = (v[:,idx]/numpy.sqrt(-w[idx])).conj().T
j2ctag = 'eig'
return j2c, j2c_negative, j2ctag
feri = h5py.File(cderi_file, 'w')
feri['j3c-kptij'] = kptij_lst
nsegs = len(fswap['j3c-junk/0'])
def make_kpt(uniq_kptji_id, cholesky_j2c): # kpt = kptj - kpti
kpt = uniq_kpts[uniq_kptji_id]
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
j2c, j2c_negative, j2ctag = cholesky_j2c
Gaux = ft_ao.ft_ao(fused_cell, Gv, None, b, gxyz, Gvbase, kpt).T
Gaux = fuse(Gaux)
Gaux *= mydf.weighted_coulG(kpt, False, mesh)
kLR = Gaux.T.real.copy('C')
kLI = Gaux.T.imag.copy('C')
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao*(nao+1)//2
if cell.dimension == 3:
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('int1e_ovlp', hermi=1, kpts=adapted_kptjs)
ovlp = [lib.pack_tril(s) for s in ovlp]
else:
aosym = 's1'
nao_pair = nao**2
mem_now = lib.current_memory()[0]
log.debug2('memory = %s', mem_now)
max_memory = max(2000, mydf.max_memory-mem_now)
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(max(int(max_memory*.38e6/16/naux/(nkptj+1)), 1), nao_pair)
shranges = _guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(16, int(max_memory*.1e6/16/buflen/(nkptj+1)))
else:
Gblksize = max(16, int(max_memory*.2e6/16/buflen/(nkptj+1)))
Gblksize = min(Gblksize, ngrids, 16384)
pqkRbuf = numpy.empty(buflen*Gblksize)
pqkIbuf = numpy.empty(buflen*Gblksize)
# buf for ft_aopair
buf = numpy.empty((nkptj,buflen*Gblksize), dtype=numpy.complex128)
def pw_contract(istep, sh_range, j3cR, j3cI):
bstart, bend, ncol = sh_range
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
else:
shls_slice = (bstart, bend, 0, cell.nbas)
for p0, p1 in lib.prange(0, ngrids, Gblksize):
dat = ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = dat[k].reshape(nG,ncol)
pqkR = numpy.ndarray((ncol,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k], 1)
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = j3cR[k]
else:
v = j3cR[k] + j3cI[k] * 1j
if j2ctag == 'CD':
v = scipy.linalg.solve_triangular(j2c, v, lower=True, overwrite_b=True)
feri['j3c/%d/%d'%(ji,istep)] = v
else:
feri['j3c/%d/%d'%(ji,istep)] = lib.dot(j2c, v)
# low-dimension systems
if j2c_negative is not None:
feri['j3c-/%d/%d'%(ji,istep)] = lib.dot(j2c_negative, v)
with lib.call_in_background(pw_contract) as compute:
col1 = 0
for istep, sh_range in enumerate(shranges):
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', \
istep+1, len(shranges), *sh_range)
bstart, bend, ncol = sh_range
col0, col1 = col1, col1+ncol
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = [fswap['j3c-junk/%d/%d'%(idx,i)][0,col0:col1].T for i in range(nsegs)]
v = fuse(numpy.vstack(v))
if is_zero(kpt) and cell.dimension == 3:
for i in numpy.where(vbar != 0)[0]:
v[i] -= vbar[i] * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
v = None
compute(istep, sh_range, j3cR, j3cI)
for ji in adapted_ji_idx:
del(fswap['j3c-junk/%d'%ji])
# Wrapped around boundary and symmetry between k and -k can be used
# explicitly for the metric integrals. We consider this symmetry
# because it is used in the df_ao2mo module when contracting two 3-index
# integral tensors to the 4-index 2e integral tensor. If the symmetry
# related k-points are treated separately, the resultant 3-index tensors
# may have inconsistent dimension due to the numerial noise when handling
# linear dependency of j2c.
def conj_j2c(cholesky_j2c):
j2c, j2c_negative, j2ctag = cholesky_j2c
if j2c_negative is None:
return j2c.conj(), None, j2ctag
else:
return j2c.conj(), j2c_negative.conj(), j2ctag
a = cell.lattice_vectors() / (2*numpy.pi)
def kconserve_indices(kpt):
'''search which (kpts+kpt) satisfies momentum conservation'''
kdif = numpy.einsum('wx,ix->wi', a, uniq_kpts + kpt)
kdif_int = numpy.rint(kdif)
mask = numpy.einsum('wi->i', abs(kdif - kdif_int)) < KPT_DIFF_TOL
uniq_kptji_ids = numpy.where(mask)[0]
return uniq_kptji_ids
done = numpy.zeros(len(uniq_kpts), dtype=bool)
for k, kpt in enumerate(uniq_kpts):
if done[k]:
continue
log.debug1('Cholesky decomposition for j2c at kpt %s', k)
cholesky_j2c = cholesky_decomposed_metric(k)
# The k-point k' which has (k - k') * a = 2n pi. Metric integrals have the
# symmetry S = S
uniq_kptji_ids = kconserve_indices(-kpt)
log.debug1("Symmetry pattern (k - %s)*a= 2n pi", kpt)
log.debug1(" make_kpt for uniq_kptji_ids %s", uniq_kptji_ids)
for uniq_kptji_id in uniq_kptji_ids:
if not done[uniq_kptji_id]:
make_kpt(uniq_kptji_id, cholesky_j2c)
done[uniq_kptji_ids] = True
# The k-point k' which has (k + k') * a = 2n pi. Metric integrals have the
# symmetry S = S*
uniq_kptji_ids = kconserve_indices(kpt)
log.debug1("Symmetry pattern (k + %s)*a= 2n pi", kpt)
log.debug1(" make_kpt for %s", uniq_kptji_ids)
cholesky_j2c = conj_j2c(cholesky_j2c)
for uniq_kptji_id in uniq_kptji_ids:
if not done[uniq_kptji_id]:
make_kpt(uniq_kptji_id, cholesky_j2c)
done[uniq_kptji_ids] = True
feri.close()
def make_kpt(kpt): # kpt = kptj - kpti
# search for all possible ki and kj that has ki-kj+kpt=0
kk_match = numpy.einsum('ijx->ij', abs(kk_table + kpt)) < 1e-9
kpti_idx, kptj_idx = numpy.where(kk_todo & kk_match)
nkptj = len(kptj_idx)
kk_todo[kpti_idx,kptj_idx] = False
if swap_2e and not is_zero(kpt):
kk_todo[kptj_idx,kpti_idx] = False
# Note: kj-ki for electorn 1 and ki-kj for electron 2
# j2c ~ ({kj-ki}|{ks-kr}) ~ ({kj-ki}|-{kj-ki}) ~ ({kj-ki}|{ki-kj})
# j3c ~ (Q|kj,ki) = j3c{ji} = (Q|ki,kj)* = conj(transpose(j3c{ij}, (0,2,1)))
bufR = numpy.empty((mydf.blockdim*nao**2))
bufI = numpy.empty((mydf.blockdim*nao**2))
for ki,kj in zip(kpti_idx,kptj_idx):
kpti = kpts_band[ki]
kptj = kpts[kj]
kptij = numpy.asarray((kpti,kptj))
for LpqR, LpqI, j3cR, j3cI in mydf.sr_loop(kptij, max_memory, False):
nrow = LpqR.shape[0]
pLqR = numpy.ndarray((nao,nrow,nao), buffer=bufR)
pLqI = numpy.ndarray((nao,nrow,nao), buffer=bufI)
pjqR = numpy.ndarray((nao,nrow,nao), buffer=LpqR)
pjqI = numpy.ndarray((nao,nrow,nao), buffer=LpqI)
tmpR = numpy.ndarray((nao,nrow*nao), buffer=j3cR)
tmpI = numpy.ndarray((nao,nrow*nao), buffer=j3cI)
pLqR[:] = LpqR.reshape(-1,nao,nao).transpose(1,0,2)
pLqI[:] = LpqI.reshape(-1,nao,nao).transpose(1,0,2)
pjqR[:] = j3cR.reshape(-1,nao,nao).transpose(1,0,2)
pjqI[:] = j3cI.reshape(-1,nao,nao).transpose(1,0,2)
#:Lpq = LpqR + LpqI*1j
#:j3c = j3cR + j3cI*1j
#:for i in range(nset):
#: dm = dms[i,ki]
#: tmp = numpy.dot(dm, j3c.reshape(nao,-1))
#: vk1 = numpy.dot(Lpq.reshape(-1,nao).conj().T, tmp.reshape(-1,nao))
#: tmp = numpy.dot(dm, Lpq.reshape(nao,-1))
#: vk1+= numpy.dot(j3c.reshape(-1,nao).conj().T, tmp.reshape(-1,nao))
#: vkR[i,kj] += vk1.real
#: vkI[i,kj] += vk1.imag
#:if swap_2e and not is_zero(kpt):
#: # K ~ 'Lij,Llk*,jk->il' + 'Llk*,Lij,jk->il'
#: for i in range(nset):
#: dm = dms[i,kj]
#: tmp = numpy.dot(j3c.reshape(-1,nao), dm)
#: vk1 = numpy.dot(tmp.reshape(nao,-1), Lpq.reshape(nao,-1).conj().T)
#: tmp = numpy.dot(Lpq.reshape(-1,nao), dm)
#: vk1+= numpy.dot(tmp.reshape(nao,-1), j3c.reshape(nao,-1).conj().T)
#: vkR[i,ki] += vk1.real
#: vkI[i,ki] += vk1.imag
# K ~ 'iLj,lLk*,li->kj' + 'lLk*,iLj,li->kj'
for i in range(nset):
tmpR, tmpI = zdotNN(dmsR[i,ki], dmsI[i,ki], pjqR.reshape(nao,-1),
pjqI.reshape(nao,-1), 1, tmpR, tmpI)
vk1R, vk1I = zdotCN(pLqR.reshape(-1,nao).T, pLqI.reshape(-1,nao).T,
tmpR.reshape(-1,nao), tmpI.reshape(-1,nao))
vkR[i,kj] += vk1R
vkI[i,kj] += vk1I
if hermi:
vkR[i,kj] += vk1R.T
vkI[i,kj] -= vk1I.T
else:
tmpR, tmpI = zdotNN(dmsR[i,ki], dmsI[i,ki], pLqR.reshape(nao,-1),
pLqI.reshape(nao,-1), 1, tmpR, tmpI)
zdotCN(pjqR.reshape(-1,nao).T, pjqI.reshape(-1,nao).T,
tmpR.reshape(-1,nao), tmpI.reshape(-1,nao),
1, vkR[i,kj], vkI[i,kj], 1)
if swap_2e and not is_zero(kpt):
tmpR = tmpR.reshape(nao*nrow,nao)
tmpI = tmpI.reshape(nao*nrow,nao)
# K ~ 'iLj,lLk*,jk->il' + 'lLk*,iLj,jk->il'
for i in range(nset):
tmpR, tmpI = zdotNN(pjqR.reshape(-1,nao), pjqI.reshape(-1,nao),
dmsR[i,kj], dmsI[i,kj], 1, tmpR, tmpI)
vk1R, vk1I = zdotNC(tmpR.reshape(nao,-1), tmpI.reshape(nao,-1),
pLqR.reshape(nao,-1).T, pLqI.reshape(nao,-1).T)
vkR[i,ki] += vk1R
vkI[i,ki] += vk1I
if hermi:
vkR[i,ki] += vk1R.T
vkI[i,ki] -= vk1I.T
else:
tmpR, tmpI = zdotNN(pLqR.reshape(-1,nao), pLqI.reshape(-1,nao),
dmsR[i,kj], dmsI[i,kj], 1, tmpR, tmpI)
zdotNC(tmpR.reshape(nao,-1), tmpI.reshape(nao,-1),
pjqR.reshape(nao,-1).T, pjqI.reshape(nao,-1).T,
1, vkR[i,ki], vkI[i,ki], 1)
LpqR = LpqI = j3cR = j3cI = tmpR = tmpI = None
return None
def _make_j3c(mydf, cell, auxcell, kptij_lst, cderi_file):
t1 = (time.clock(), time.time())
log = logger.Logger(mydf.stdout, mydf.verbose)
max_memory = max(2000, mydf.max_memory - lib.current_memory()[0])
fused_cell, fuse = fuse_auxcell(mydf, auxcell)
# The ideal way to hold the temporary integrals is to store them in the
# cderi_file and overwrite them inplace in the second pass. The current
# HDF5 library does not have an efficient way to manage free space in
# overwriting. It often leads to the cderi_file ~2 times larger than the
# necessary size. For now, dumping the DF integral intermediates to a
# separated temporary file can avoid this issue. The DF intermediates may
# be terribly huge. The temporary file should be placed in the same disk
# as cderi_file.
swapfile = tempfile.NamedTemporaryFile(dir=os.path.dirname(cderi_file))
fswap = lib.H5TmpFile(swapfile.name)
# Unlink swapfile to avoid trash
swapfile = None
outcore._aux_e2(cell,
fused_cell,
fswap,
'int3c2e',
aosym='s2',
kptij_lst=kptij_lst,
dataname='j3c-junk',
max_memory=max_memory)
t1 = log.timer_debug1('3c2e', *t1)
nao = cell.nao_nr()
naux = auxcell.nao_nr()
mesh = mydf.mesh
Gv, Gvbase, kws = cell.get_Gv_weights(mesh)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngrids = gxyz.shape[0]
kptis = kptij_lst[:, 0]
kptjs = kptij_lst[:, 1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
log.debug('Num uniq kpts %d', len(uniq_kpts))
log.debug2('uniq_kpts %s', uniq_kpts)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('int2c2e', hermi=1, kpts=uniq_kpts)
max_memory = max(2000, mydf.max_memory - lib.current_memory()[0])
blksize = max(2048, int(max_memory * .5e6 / 16 / fused_cell.nao_nr()))
log.debug2('max_memory %s (MB) blocksize %s', max_memory, blksize)
for k, kpt in enumerate(uniq_kpts):
coulG = mydf.weighted_coulG(kpt, False, mesh)
for p0, p1 in lib.prange(0, ngrids, blksize):
aoaux = ft_ao.ft_ao(fused_cell, Gv[p0:p1], None, b, gxyz[p0:p1],
Gvbase, kpt).T
LkR = numpy.asarray(aoaux.real, order='C')
LkI = numpy.asarray(aoaux.imag, order='C')
aoaux = None
if is_zero(kpt): # kpti == kptj
j2c[k][naux:] -= lib.ddot(LkR[naux:] * coulG[p0:p1], LkR.T)
j2c[k][naux:] -= lib.ddot(LkI[naux:] * coulG[p0:p1], LkI.T)
j2c[k][:naux, naux:] = j2c[k][naux:, :naux].T
else:
j2cR, j2cI = zdotCN(LkR[naux:] * coulG[p0:p1],
LkI[naux:] * coulG[p0:p1], LkR.T, LkI.T)
j2c[k][naux:] -= j2cR + j2cI * 1j
j2c[k][:naux, naux:] = j2c[k][naux:, :naux].T.conj()
LkR = LkI = None
fswap['j2c/%d' % k] = fuse(fuse(j2c[k]).T).T
j2c = coulG = None
def cholesky_decomposed_metric(uniq_kptji_id):
j2c = numpy.asarray(fswap['j2c/%d' % uniq_kptji_id])
j2c_negative = None
try:
j2c = scipy.linalg.cholesky(j2c, lower=True)
j2ctag = 'CD'
except scipy.linalg.LinAlgError:
#msg =('===================================\n'
# 'J-metric not positive definite.\n'
# 'It is likely that mesh is not enough.\n'
# '===================================')
#log.error(msg)
#raise scipy.linalg.LinAlgError('\n'.join([str(e), msg]))
w, v = scipy.linalg.eigh(j2c)
log.debug('DF metric linear dependency for kpt %s', uniq_kptji_id)
log.debug('cond = %.4g, drop %d bfns', w[-1] / w[0],
numpy.count_nonzero(w < mydf.linear_dep_threshold))
v1 = v[:, w > mydf.linear_dep_threshold].conj().T
v1 /= numpy.sqrt(w[w > mydf.linear_dep_threshold]).reshape(-1, 1)
j2c = v1
if cell.dimension == 2 and cell.low_dim_ft_type != 'inf_vacuum':
idx = numpy.where(w < -mydf.linear_dep_threshold)[0]
if len(idx) > 0:
j2c_negative = (v[:, idx] / numpy.sqrt(-w[idx])).conj().T
w = v = None
j2ctag = 'eig'
return j2c, j2c_negative, j2ctag
feri = h5py.File(cderi_file, 'w')
feri['j3c-kptij'] = kptij_lst
nsegs = len(fswap['j3c-junk/0'])
def make_kpt(uniq_kptji_id, cholesky_j2c):
kpt = uniq_kpts[uniq_kptji_id] # kpt = kptj - kpti
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
j2c, j2c_negative, j2ctag = cholesky_j2c
shls_slice = (auxcell.nbas, fused_cell.nbas)
Gaux = ft_ao.ft_ao(fused_cell, Gv, shls_slice, b, gxyz, Gvbase, kpt)
wcoulG = mydf.weighted_coulG(kpt, False, mesh)
Gaux *= wcoulG.reshape(-1, 1)
kLR = Gaux.real.copy('C')
kLI = Gaux.imag.copy('C')
Gaux = None
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao * (nao + 1) // 2
if cell.dimension == 3:
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('int1e_ovlp',
hermi=1,
kpts=adapted_kptjs)
ovlp = [lib.pack_tril(s) for s in ovlp]
else:
aosym = 's1'
nao_pair = nao**2
mem_now = lib.current_memory()[0]
log.debug2('memory = %s', mem_now)
max_memory = max(2000, mydf.max_memory - mem_now)
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(max(int(max_memory * .38e6 / 16 / naux / (nkptj + 1)), 1),
nao_pair)
shranges = _guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(16,
int(max_memory * .1e6 / 16 / buflen / (nkptj + 1)))
else:
Gblksize = max(16,
int(max_memory * .2e6 / 16 / buflen / (nkptj + 1)))
Gblksize = min(Gblksize, ngrids, 16384)
def load(aux_slice):
col0, col1 = aux_slice
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = numpy.vstack([
fswap['j3c-junk/%d/%d' % (idx, i)][0, col0:col1].T
for i in range(nsegs)
])
# vbar is the interaction between the background charge
# and the auxiliary basis. 0D, 1D, 2D do not have vbar.
if is_zero(kpt) and cell.dimension == 3:
for i in numpy.where(vbar != 0)[0]:
v[i] -= vbar[i] * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
v = None
return j3cR, j3cI
pqkRbuf = numpy.empty(buflen * Gblksize)
pqkIbuf = numpy.empty(buflen * Gblksize)
# buf for ft_aopair
buf = numpy.empty(nkptj * buflen * Gblksize, dtype=numpy.complex128)
cols = [sh_range[2] for sh_range in shranges]
locs = numpy.append(0, numpy.cumsum(cols))
tasks = zip(locs[:-1], locs[1:])
for istep, (j3cR,
j3cI) in enumerate(lib.map_with_prefetch(load, tasks)):
bstart, bend, ncol = shranges[istep]
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', istep + 1,
len(shranges), bstart, bend, ncol)
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
else:
shls_slice = (bstart, bend, 0, cell.nbas)
for p0, p1 in lib.prange(0, ngrids, Gblksize):
dat = ft_ao._ft_aopair_kpts(cell,
Gv[p0:p1],
shls_slice,
aosym,
b,
gxyz[p0:p1],
Gvbase,
kpt,
adapted_kptjs,
out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = dat[k].reshape(nG, ncol)
pqkR = numpy.ndarray((ncol, nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol, nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k][naux:], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k][naux:], 1)
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = fuse(j3cR[k])
else:
v = fuse(j3cR[k] + j3cI[k] * 1j)
if j2ctag == 'CD':
v = scipy.linalg.solve_triangular(j2c,
v,
lower=True,
overwrite_b=True)
feri['j3c/%d/%d' % (ji, istep)] = v
else:
feri['j3c/%d/%d' % (ji, istep)] = lib.dot(j2c, v)
# low-dimension systems
if j2c_negative is not None:
feri['j3c-/%d/%d' % (ji, istep)] = lib.dot(j2c_negative, v)
j3cR = j3cI = None
for ji in adapted_ji_idx:
del (fswap['j3c-junk/%d' % ji])
# Wrapped around boundary and symmetry between k and -k can be used
# explicitly for the metric integrals. We consider this symmetry
# because it is used in the df_ao2mo module when contracting two 3-index
# integral tensors to the 4-index 2e integral tensor. If the symmetry
# related k-points are treated separately, the resultant 3-index tensors
# may have inconsistent dimension due to the numerial noise when handling
# linear dependency of j2c.
def conj_j2c(cholesky_j2c):
j2c, j2c_negative, j2ctag = cholesky_j2c
if j2c_negative is None:
return j2c.conj(), None, j2ctag
else:
return j2c.conj(), j2c_negative.conj(), j2ctag
a = cell.lattice_vectors() / (2 * numpy.pi)
def kconserve_indices(kpt):
'''search which (kpts+kpt) satisfies momentum conservation'''
kdif = numpy.einsum('wx,ix->wi', a, uniq_kpts + kpt)
kdif_int = numpy.rint(kdif)
mask = numpy.einsum('wi->i', abs(kdif - kdif_int)) < KPT_DIFF_TOL
uniq_kptji_ids = numpy.where(mask)[0]
return uniq_kptji_ids
done = numpy.zeros(len(uniq_kpts), dtype=bool)
for k, kpt in enumerate(uniq_kpts):
if done[k]:
continue
log.debug1('Cholesky decomposition for j2c at kpt %s', k)
cholesky_j2c = cholesky_decomposed_metric(k)
# The k-point k' which has (k - k') * a = 2n pi. Metric integrals have the
# symmetry S = S
uniq_kptji_ids = kconserve_indices(-kpt)
log.debug1("Symmetry pattern (k - %s)*a= 2n pi", kpt)
log.debug1(" make_kpt for uniq_kptji_ids %s", uniq_kptji_ids)
for uniq_kptji_id in uniq_kptji_ids:
if not done[uniq_kptji_id]:
make_kpt(uniq_kptji_id, cholesky_j2c)
done[uniq_kptji_ids] = True
# The k-point k' which has (k + k') * a = 2n pi. Metric integrals have the
# symmetry S = S*
uniq_kptji_ids = kconserve_indices(kpt)
log.debug1("Symmetry pattern (k + %s)*a= 2n pi", kpt)
log.debug1(" make_kpt for %s", uniq_kptji_ids)
cholesky_j2c = conj_j2c(cholesky_j2c)
for uniq_kptji_id in uniq_kptji_ids:
if not done[uniq_kptji_id]:
make_kpt(uniq_kptji_id, cholesky_j2c)
done[uniq_kptji_ids] = True
feri.close()
def _make_j3c(mydf, cell, auxcell, kptij_lst, cderi_file):
log = logger.Logger(mydf.stdout, mydf.verbose)
t1 = t0 = (time.clock(), time.time())
fused_cell, fuse = fuse_auxcell(mydf, mydf.auxcell)
ao_loc = cell.ao_loc_nr()
nao = ao_loc[-1]
naux = auxcell.nao_nr()
nkptij = len(kptij_lst)
gs = mydf.gs
Gv, Gvbase, kws = cell.get_Gv_weights(gs)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngs = gxyz.shape[0]
kptis = kptij_lst[:, 0]
kptjs = kptij_lst[:, 1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
log.debug('Num uniq kpts %d', len(uniq_kpts))
log.debug2('uniq_kpts %s', uniq_kpts)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('int2c2e_sph', hermi=1, kpts=uniq_kpts)
j2ctags = []
nauxs = []
t1 = log.timer_debug1('2c2e', *t1)
if h5py.is_hdf5(cderi_file):
feri = h5py.File(cderi_file)
else:
feri = h5py.File(cderi_file, 'w')
for k, kpt in enumerate(uniq_kpts):
aoaux = ft_ao.ft_ao(fused_cell, Gv, None, b, gxyz, Gvbase, kpt).T
coulG = numpy.sqrt(mydf.weighted_coulG(kpt, False, gs))
kLR = (aoaux.real * coulG).T
kLI = (aoaux.imag * coulG).T
if not kLR.flags.c_contiguous: kLR = lib.transpose(kLR.T)
if not kLI.flags.c_contiguous: kLI = lib.transpose(kLI.T)
aoaux = None
kLR1 = numpy.asarray(kLR[:, naux:], order='C')
kLI1 = numpy.asarray(kLI[:, naux:], order='C')
if is_zero(kpt): # kpti == kptj
for p0, p1 in mydf.mpi_prange(0, ngs):
j2cR = lib.ddot(kLR1[p0:p1].T, kLR[p0:p1])
j2cR = lib.ddot(kLI1[p0:p1].T, kLI[p0:p1], 1, j2cR, 1)
j2c[k][naux:] -= mpi.allreduce(j2cR)
j2c[k][:naux, naux:] = j2c[k][naux:, :naux].T
else:
for p0, p1 in mydf.mpi_prange(0, ngs):
j2cR, j2cI = zdotCN(kLR1[p0:p1].T, kLI1[p0:p1].T, kLR[p0:p1],
kLI[p0:p1])
j2cR = mpi.allreduce(j2cR)
j2cI = mpi.allreduce(j2cI)
j2c[k][naux:] -= j2cR + j2cI * 1j
j2c[k][:naux, naux:] = j2c[k][naux:, :naux].T.conj()
j2c[k] = fuse(fuse(j2c[k]).T).T
try:
feri['j2c/%d' % k] = scipy.linalg.cholesky(j2c[k], lower=True)
j2ctags.append('CD')
nauxs.append(naux)
except scipy.linalg.LinAlgError as e:
#msg =('===================================\n'
# 'J-metric not positive definite.\n'
# 'It is likely that gs is not enough.\n'
# '===================================')
#log.error(msg)
#raise scipy.linalg.LinAlgError('\n'.join([e.message, msg]))
w, v = scipy.linalg.eigh(j2c)
log.debug2('metric linear dependency for kpt %s', uniq_kptji_id)
log.debug2('cond = %.4g, drop %d bfns', w[0] / w[-1],
numpy.count_nonzero(w < LINEAR_DEP_THR))
v = v[:, w > LINEAR_DEP_THR].T.conj()
v /= numpy.sqrt(w[w > LINEAR_DEP_THR]).reshape(-1, 1)
feri['j2c/%d' % k] = v
j2ctags.append('eig')
nauxs.append(v.shape[0])
kLR = kLI = kLR1 = kLI1 = coulG = None
j2c = None
aosym_s2 = numpy.einsum('ix->i', abs(kptis - kptjs)) < 1e-9
j_only = numpy.all(aosym_s2)
if gamma_point(kptij_lst):
dtype = 'f8'
else:
dtype = 'c16'
vbar = mydf.auxbar(fused_cell)
vbar = fuse(vbar)
ovlp = cell.pbc_intor('int1e_ovlp_sph', hermi=1, kpts=kptjs[aosym_s2])
ovlp = [lib.pack_tril(s) for s in ovlp]
t1 = log.timer_debug1('aoaux and int2c', *t1)
# Estimates the buffer size based on the last contraction in G-space.
# This contraction requires to hold nkptj copies of (naux,?) array
# simultaneously in memory.
mem_now = max(comm.allgather(lib.current_memory()[0]))
max_memory = max(2000, mydf.max_memory - mem_now)
nkptj_max = max((uniq_inverse == x).sum() for x in set(uniq_inverse))
buflen = max(
int(
min(max_memory * .5e6 / 16 / naux / (nkptj_max + 2) / nao,
nao / 3 / mpi.pool.size)), 1)
chunks = (buflen, nao)
j3c_jobs = grids2d_int3c_jobs(cell, auxcell, kptij_lst, chunks, j_only)
log.debug1('max_memory = %d MB (%d in use) chunks %s', max_memory,
mem_now, chunks)
log.debug2('j3c_jobs %s', j3c_jobs)
if j_only:
int3c = wrap_int3c(cell, fused_cell, 'int3c2e_sph', 's2', 1, kptij_lst)
else:
int3c = wrap_int3c(cell, fused_cell, 'int3c2e_sph', 's1', 1, kptij_lst)
idxb = numpy.tril_indices(nao)
idxb = (idxb[0] * nao + idxb[1]).astype('i')
aux_loc = fused_cell.ao_loc_nr('ssc' in 'int3c2e_sph')
def gen_int3c(auxcell, job_id, ish0, ish1):
dataname = 'j3c-chunks/%d' % job_id
if dataname in feri:
del (feri[dataname])
i0 = ao_loc[ish0]
i1 = ao_loc[ish1]
dii = i1 * (i1 + 1) // 2 - i0 * (i0 + 1) // 2
dij = (i1 - i0) * nao
if j_only:
buflen = max(8, int(max_memory * 1e6 / 16 / (nkptij * dii + dii)))
else:
buflen = max(8, int(max_memory * 1e6 / 16 / (nkptij * dij + dij)))
auxranges = balance_segs(aux_loc[1:] - aux_loc[:-1], buflen)
buflen = max([x[2] for x in auxranges])
buf = numpy.empty(nkptij * dij * buflen, dtype=dtype)
buf1 = numpy.empty(dij * buflen, dtype=dtype)
naux = aux_loc[-1]
for kpt_id, kptij in enumerate(kptij_lst):
key = '%s/%d' % (dataname, kpt_id)
if aosym_s2[kpt_id]:
shape = (naux, dii)
else:
shape = (naux, dij)
if gamma_point(kptij):
feri.create_dataset(key, shape, 'f8')
else:
feri.create_dataset(key, shape, 'c16')
naux0 = 0
for istep, auxrange in enumerate(auxranges):
log.alldebug2("aux_e2 job_id %d step %d", job_id, istep)
sh0, sh1, nrow = auxrange
sub_slice = (ish0, ish1, 0, cell.nbas, sh0, sh1)
if j_only:
mat = numpy.ndarray((nkptij, dii, nrow),
dtype=dtype,
buffer=buf)
else:
mat = numpy.ndarray((nkptij, dij, nrow),
dtype=dtype,
buffer=buf)
mat = int3c(sub_slice, mat)
for k, kptij in enumerate(kptij_lst):
h5dat = feri['%s/%d' % (dataname, k)]
v = lib.transpose(mat[k], out=buf1)
if not j_only and aosym_s2[k]:
idy = idxb[i0 * (i0 + 1) // 2:i1 *
(i1 + 1) // 2] - i0 * nao
out = numpy.ndarray((nrow, dii),
dtype=v.dtype,
buffer=mat[k])
v = numpy.take(v, idy, axis=1, out=out)
if gamma_point(kptij):
h5dat[naux0:naux0 + nrow] = v.real
else:
h5dat[naux0:naux0 + nrow] = v
naux0 += nrow
def ft_fuse(job_id, uniq_kptji_id, sh0, sh1):
kpt = uniq_kpts[uniq_kptji_id] # kpt = kptj - kpti
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
shls_slice = (auxcell.nbas, fused_cell.nbas)
Gaux = ft_ao.ft_ao(fused_cell, Gv, shls_slice, b, gxyz, Gvbase, kpt)
Gaux *= mydf.weighted_coulG(kpt, False, gs).reshape(-1, 1)
kLR = Gaux.real.copy('C')
kLI = Gaux.imag.copy('C')
j2c = numpy.asarray(feri['j2c/%d' % uniq_kptji_id])
j2ctag = j2ctags[uniq_kptji_id]
naux0 = j2c.shape[0]
if is_zero(kpt):
aosym = 's2'
else:
aosym = 's1'
j3cR = [None] * nkptj
j3cI = [None] * nkptj
i0 = ao_loc[sh0]
i1 = ao_loc[sh1]
for k, idx in enumerate(adapted_ji_idx):
key = 'j3c-chunks/%d/%d' % (job_id, idx)
v = numpy.asarray(feri[key])
if is_zero(kpt):
for i, c in enumerate(vbar):
if c != 0:
v[i] -= c * ovlp[k][i0 * (i0 + 1) // 2:i1 *
(i1 + 1) // 2].ravel()
j3cR[k] = numpy.asarray(v.real, order='C')
if v.dtype == numpy.complex128:
j3cI[k] = numpy.asarray(v.imag, order='C')
v = None
ncol = j3cR[0].shape[1]
Gblksize = max(16, int(max_memory * 1e6 / 16 / ncol /
(nkptj + 1))) # +1 for pqkRbuf/pqkIbuf
Gblksize = min(Gblksize, ngs, 16384)
pqkRbuf = numpy.empty(ncol * Gblksize)
pqkIbuf = numpy.empty(ncol * Gblksize)
buf = numpy.empty(nkptj * ncol * Gblksize, dtype=numpy.complex128)
log.alldebug2(' blksize (%d,%d)', Gblksize, ncol)
shls_slice = (sh0, sh1, 0, cell.nbas)
for p0, p1 in lib.prange(0, ngs, Gblksize):
dat = ft_ao._ft_aopair_kpts(cell,
Gv[p0:p1],
shls_slice,
aosym,
b,
gxyz[p0:p1],
Gvbase,
kpt,
adapted_kptjs,
out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = dat[k].reshape(nG, ncol)
pqkR = numpy.ndarray((ncol, nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol, nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k][naux:], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k][naux:], 1)
for k, idx in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = fuse(j3cR[k])
else:
v = fuse(j3cR[k] + j3cI[k] * 1j)
if j2ctag == 'CD':
v = scipy.linalg.solve_triangular(j2c,
v,
lower=True,
overwrite_b=True)
else:
v = lib.dot(j2c, v)
feri['j3c-chunks/%d/%d' % (job_id, idx)][:naux0] = v
t2 = t1
j3c_workers = numpy.zeros(len(j3c_jobs), dtype=int)
#for job_id, ish0, ish1 in mpi.work_share_partition(j3c_jobs):
for job_id, ish0, ish1 in mpi.work_stealing_partition(j3c_jobs):
gen_int3c(fused_cell, job_id, ish0, ish1)
t2 = log.alltimer_debug2('int j3c %d' % job_id, *t2)
for k, kpt in enumerate(uniq_kpts):
ft_fuse(job_id, k, ish0, ish1)
t2 = log.alltimer_debug2('ft-fuse %d k %d' % (job_id, k), *t2)
j3c_workers[job_id] = rank
j3c_workers = mpi.allreduce(j3c_workers)
log.debug2('j3c_workers %s', j3c_workers)
j2c = kLRs = kLIs = ovlp = vbar = fuse = gen_int3c = ft_fuse = None
t1 = log.timer_debug1('int3c and fuse', *t1)
def get_segs_loc(aosym):
off0 = numpy.asarray([ao_loc[i0] for x, i0, i1 in j3c_jobs])
off1 = numpy.asarray([ao_loc[i1] for x, i0, i1 in j3c_jobs])
if aosym: # s2
dims = off1 * (off1 + 1) // 2 - off0 * (off0 + 1) // 2
else:
dims = (off1 - off0) * nao
#dims = numpy.asarray([ao_loc[i1]-ao_loc[i0] for x,i0,i1 in j3c_jobs])
dims = numpy.hstack(
[dims[j3c_workers == w] for w in range(mpi.pool.size)])
job_idx = numpy.hstack(
[numpy.where(j3c_workers == w)[0] for w in range(mpi.pool.size)])
segs_loc = numpy.append(0, numpy.cumsum(dims))
segs_loc = [(segs_loc[j], segs_loc[j + 1])
for j in numpy.argsort(job_idx)]
return segs_loc
segs_loc_s1 = get_segs_loc(False)
segs_loc_s2 = get_segs_loc(True)
if 'j3c' in feri: del (feri['j3c'])
segsize = (max(nauxs) + mpi.pool.size - 1) // mpi.pool.size
naux0 = rank * segsize
for k, kptij in enumerate(kptij_lst):
naux1 = min(nauxs[uniq_inverse[k]], naux0 + segsize)
nrow = max(0, naux1 - naux0)
if gamma_point(kptij):
dtype = 'f8'
else:
dtype = 'c16'
if aosym_s2[k]:
nao_pair = nao * (nao + 1) // 2
else:
nao_pair = nao * nao
feri.create_dataset('j3c/%d' % k, (nrow, nao_pair),
dtype,
maxshape=(None, nao_pair))
def load(k, p0, p1):
naux1 = nauxs[uniq_inverse[k]]
slices = [(min(i * segsize + p0, naux1), min(i * segsize + p1, naux1))
for i in range(mpi.pool.size)]
segs = []
for p0, p1 in slices:
val = []
for job_id, worker in enumerate(j3c_workers):
if rank == worker:
key = 'j3c-chunks/%d/%d' % (job_id, k)
val.append(feri[key][p0:p1].ravel())
if val:
segs.append(numpy.hstack(val))
else:
segs.append(numpy.zeros(0))
return segs
def save(k, p0, p1, segs):
segs = mpi.alltoall(segs)
naux1 = nauxs[uniq_inverse[k]]
loc0, loc1 = min(p0, naux1 - naux0), min(p1, naux1 - naux0)
nL = loc1 - loc0
if nL > 0:
if aosym_s2[k]:
segs = numpy.hstack([
segs[i0 * nL:i1 * nL].reshape(nL, -1)
for i0, i1 in segs_loc_s2
])
else:
segs = numpy.hstack([
segs[i0 * nL:i1 * nL].reshape(nL, -1)
for i0, i1 in segs_loc_s1
])
feri['j3c/%d' % k][loc0:loc1] = segs
mem_now = max(comm.allgather(lib.current_memory()[0]))
max_memory = max(2000, min(8000, mydf.max_memory - mem_now))
if numpy.all(aosym_s2):
if gamma_point(kptij_lst):
blksize = max(16, int(max_memory * .5e6 / 8 / nao**2))
else:
blksize = max(16, int(max_memory * .5e6 / 16 / nao**2))
else:
blksize = max(16, int(max_memory * .5e6 / 16 / nao**2 / 2))
log.debug1('max_momory %d MB (%d in use), blksize %d', max_memory, mem_now,
blksize)
t2 = t1
with lib.call_in_background(save) as async_write:
for k, kptji in enumerate(kptij_lst):
for p0, p1 in lib.prange(0, segsize, blksize):
segs = load(k, p0, p1)
async_write(k, p0, p1, segs)
t2 = log.timer_debug1(
'assemble k=%d %d:%d (in %d)' % (k, p0, p1, segsize), *t2)
if 'j3c-chunks' in feri: del (feri['j3c-chunks'])
if 'j3c-kptij' in feri: del (feri['j3c-kptij'])
feri['j3c-kptij'] = kptij_lst
t1 = log.alltimer_debug1('assembling j3c', *t1)
feri.close()
def _make_j3c(mydf, cell, auxcell, kptij_lst):
t1 = (time.clock(), time.time())
log = logger.Logger(mydf.stdout, mydf.verbose)
max_memory = max(2000, mydf.max_memory-lib.current_memory()[0])
fused_cell, fuse = fuse_auxcell(mydf, auxcell)
outcore.aux_e2(cell, fused_cell, mydf._cderi, 'cint3c2e_sph',
kptij_lst=kptij_lst, dataname='j3c', max_memory=max_memory)
t1 = log.timer_debug1('3c2e', *t1)
nao = cell.nao_nr()
naux = auxcell.nao_nr()
gs = mydf.gs
Gv, Gvbase, kws = cell.get_Gv_weights(gs)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngs = gxyz.shape[0]
kptis = kptij_lst[:,0]
kptjs = kptij_lst[:,1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('cint2c2e_sph', hermi=1, kpts=uniq_kpts)
kLRs = []
kLIs = []
# chgcell = make_modchg_basis(auxcell, mydf.eta)
# for k, kpt in enumerate(uniq_kpts):
# aoaux = ft_ao.ft_ao(chgcell, Gv, None, b, gxyz, Gvbase, kpt).T
# coulG = numpy.sqrt(mydf.weighted_coulG(kpt, False, gs))
# LkR = aoaux.real * coulG
# LkI = aoaux.imag * coulG
# j2caux = numpy.zeros_like(j2c[k])
# j2caux[naux:,naux:] = j2c[naux:,naux:]
# if is_zero(kpt): # kpti == kptj
# j2caux[naux:,naux:] -= lib.ddot(LkR, LkR.T)
# j2caux[naux:,naux:] -= lib.ddot(LkI, LkI.T)
# j2c[k] = j2c[k][:naux,:naux] - fuse(fuse(j2caux.T).T)
# vbar = fuse(mydf.auxbar(fused_cell))
# s = (vbar != 0).astype(numpy.double)
# j2c[k] -= numpy.einsum('i,j->ij', vbar, s)
# j2c[k] -= numpy.einsum('i,j->ij', s, vbar)
# else:
# j2cR, j2cI = zdotCN(LkR, LkI, LkR.T, LkI.T)
# j2caux[naux:,naux:] -= j2cR + j2cI * 1j
# j2c[k] = j2c[k][:naux,:naux] - fuse(fuse(j2caux.T).T)
# #j2c[k] = fuse(fuse(j2c[k]).T).T.copy()
# try:
# j2c[k] = scipy.linalg.cholesky(fuse(fuse(j2c[k]).T).T, lower=True)
# except scipy.linalg.LinAlgError as e:
# msg =('===================================\n'
# 'J-metric not positive definite.\n'
# 'It is likely that gs is not enough.\n'
# '===================================')
# log.error(msg)
# raise scipy.linalg.LinAlgError('\n'.join([e.message, msg]))
# kLR = LkR.T
# kLI = LkI.T
# if not kLR.flags.c_contiguous: kLR = lib.transpose(LkR)
# if not kLI.flags.c_contiguous: kLI = lib.transpose(LkI)
# kLR *= coulG.reshape(-1,1)
# kLI *= coulG.reshape(-1,1)
# kLRs.append(kLR)
# kLIs.append(kLI)
# aoaux = LkR = LkI = kLR = kLI = coulG = None
for k, kpt in enumerate(uniq_kpts):
aoaux = ft_ao.ft_ao(fused_cell, Gv, None, b, gxyz, Gvbase, kpt).T
coulG = numpy.sqrt(mydf.weighted_coulG(kpt, False, gs))
LkR = aoaux.real * coulG
LkI = aoaux.imag * coulG
if is_zero(kpt): # kpti == kptj
j2c[k][naux:] -= lib.ddot(LkR[naux:], LkR.T)
j2c[k][naux:] -= lib.ddot(LkI[naux:], LkI.T)
j2c[k][:naux,naux:] = j2c[k][naux:,:naux].T
else:
j2cR, j2cI = zdotCN(LkR[naux:], LkI[naux:], LkR.T, LkI.T)
j2c[k][naux:] -= j2cR + j2cI * 1j
j2c[k][:naux,naux:] = j2c[k][naux:,:naux].T.conj()
#j2c[k] = fuse(fuse(j2c[k]).T).T.copy()
try:
j2c[k] = scipy.linalg.cholesky(fuse(fuse(j2c[k]).T).T, lower=True)
except scipy.linalg.LinAlgError as e:
msg =('===================================\n'
'J-metric not positive definite.\n'
'It is likely that gs is not enough.\n'
'===================================')
log.error(msg)
raise scipy.linalg.LinAlgError('\n'.join([e.message, msg]))
kLR = LkR[naux:].T
kLI = LkI[naux:].T
if not kLR.flags.c_contiguous: kLR = lib.transpose(LkR[naux:])
if not kLI.flags.c_contiguous: kLI = lib.transpose(LkI[naux:])
kLR *= coulG.reshape(-1,1)
kLI *= coulG.reshape(-1,1)
kLRs.append(kLR)
kLIs.append(kLI)
aoaux = LkR = LkI = kLR = kLI = coulG = None
feri = h5py.File(mydf._cderi)
def make_kpt(uniq_kptji_id): # kpt = kptj - kpti
kpt = uniq_kpts[uniq_kptji_id]
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
kLR = kLRs[uniq_kptji_id]
kLI = kLIs[uniq_kptji_id]
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao*(nao+1)//2
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('cint1e_ovlp_sph', hermi=1, kpts=adapted_kptjs)
for k, ji in enumerate(adapted_ji_idx):
ovlp[k] = lib.pack_tril(ovlp[k])
else:
aosym = 's1'
nao_pair = nao**2
max_memory = max(2000, mydf.max_memory-lib.current_memory()[0])
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(max(int(max_memory*.6*1e6/16/naux/(nkptj+1)), 1), nao_pair)
shranges = _guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(16, int(max_memory*.2*1e6/16/buflen/(nkptj+1)))
else:
Gblksize = max(16, int(max_memory*.4*1e6/16/buflen/(nkptj+1)))
Gblksize = min(Gblksize, ngs, 16384)
pqkRbuf = numpy.empty(buflen*Gblksize)
pqkIbuf = numpy.empty(buflen*Gblksize)
# buf for ft_aopair
buf = numpy.zeros((nkptj,buflen*Gblksize), dtype=numpy.complex128)
col1 = 0
for istep, sh_range in enumerate(shranges):
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', \
istep+1, len(shranges), *sh_range)
bstart, bend, ncol = sh_range
col0, col1 = col1, col1+ncol
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = numpy.asarray(feri['j3c/%d'%idx][:,col0:col1])
if is_zero(kpt):
for i, c in enumerate(vbar):
if c != 0:
v[i] -= c * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG,ncol), dtype=numpy.complex128,
order='F', buffer=buf[k])
pqkR = numpy.ndarray((ncol,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
aoao[:] = 0
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k][naux:], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k][naux:], 1)
else:
shls_slice = (bstart, bend, 0, cell.nbas)
ni = ncol // nao
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG,ni,nao), dtype=numpy.complex128,
order='F', buffer=buf[k])
pqkR = numpy.ndarray((ni,nao,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ni,nao,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.transpose(1,2,0)
pqkI[:] = aoao.imag.transpose(1,2,0)
aoao[:] = 0
pqkR = pqkR.reshape(-1,nG)
pqkI = pqkI.reshape(-1,nG)
zdotCN(kLR[p0:p1].T, kLI[p0:p1].T, pqkR.T, pqkI.T,
-1, j3cR[k][naux:], j3cI[k][naux:], 1)
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = fuse(j3cR[k])
else:
v = fuse(j3cR[k] + j3cI[k] * 1j)
v = scipy.linalg.solve_triangular(j2c[uniq_kptji_id], v,
lower=True, overwrite_b=True)
feri['j3c/%d'%ji][:naux,col0:col1] = v
for k, kpt in enumerate(uniq_kpts):
make_kpt(k)
for k, kptij in enumerate(kptij_lst):
v = feri['j3c/%d'%k][:naux]
del(feri['j3c/%d'%k])
feri['j3c/%d'%k] = v
feri.close()
def make_kpt(kpt): # kpt = kptj - kpti
# search for all possible ki and kj that has ki-kj+kpt=0
kk_match = numpy.einsum('ijx->ij', abs(kk_table + kpt)) < 1e-9
kpti_idx, kptj_idx = numpy.where(kk_todo & kk_match)
nkptj = len(kptj_idx)
log.debug1('kpt = %s', kpt)
log.debug2('kpti_idx = %s', kpti_idx)
log.debug2('kptj_idx = %s', kptj_idx)
kk_todo[kpti_idx,kptj_idx] = False
if swap_2e and not is_zero(kpt):
kk_todo[kptj_idx,kpti_idx] = False
max_memory1 = max_memory * (nkptj+1)/(nkptj+5)
#blksize = max(int(max_memory1*4e6/(nkptj+5)/16/nao**2), 16)
#bufR = numpy.empty((blksize*nao**2))
#bufI = numpy.empty((blksize*nao**2))
# Use DF object to mimic KRHF/KUHF object in function get_coulG
mydf.exxdiv = exxdiv
vkcoulG = mydf.weighted_coulG(kpt, True, mesh)
kptjs = kpts[kptj_idx]
# <r|-G+k_rs|s> = conj(<s|G-k_rs|r>) = conj(<s|G+k_sr|r>)
#buf1R = numpy.empty((blksize*nao**2))
#buf1I = numpy.empty((blksize*nao**2))
for aoaoks, p0, p1 in mydf.ft_loop(mesh, kpt, kptjs, max_memory=max_memory1):
nG = p1 - p0
bufR = numpy.empty((nG*nao**2))
bufI = numpy.empty((nG*nao**2))
buf1R = numpy.empty((nG*nao**2))
buf1I = numpy.empty((nG*nao**2))
for k, aoao in enumerate(aoaoks):
ki = kpti_idx[k]
kj = kptj_idx[k]
# case 1: k_pq = (pi|iq)
#:v4 = numpy.einsum('ijL,lkL->ijkl', pqk, pqk.conj())
#:vk += numpy.einsum('ijkl,jk->il', v4, dm)
pLqR = numpy.ndarray((nao,nG,nao), buffer=bufR)
pLqI = numpy.ndarray((nao,nG,nao), buffer=bufI)
pLqR[:] = aoao.real.reshape(nG,nao,nao).transpose(1,0,2)
pLqI[:] = aoao.imag.reshape(nG,nao,nao).transpose(1,0,2)
iLkR = numpy.ndarray((nao,nG,nao), buffer=buf1R)
iLkI = numpy.ndarray((nao,nG,nao), buffer=buf1I)
for i in range(nset):
zdotNN(pLqR.reshape(-1,nao), pLqI.reshape(-1,nao),
dmsR[i,kj], dmsI[i,kj], 1,
iLkR.reshape(-1,nao), iLkI.reshape(-1,nao))
iLkR *= vkcoulG[p0:p1].reshape(1,nG,1)
iLkI *= vkcoulG[p0:p1].reshape(1,nG,1)
zdotNC(iLkR.reshape(nao,-1), iLkI.reshape(nao,-1),
pLqR.reshape(nao,-1).T, pLqI.reshape(nao,-1).T,
1, vkR[i,ki], vkI[i,ki], 1)
# case 2: k_pq = (iq|pi)
#:v4 = numpy.einsum('iLj,lLk->ijkl', pqk, pqk.conj())
#:vk += numpy.einsum('ijkl,li->kj', v4, dm)
if swap_2e and not is_zero(kpt):
for i in range(nset):
zdotNN(dmsR[i,ki], dmsI[i,ki], pLqR.reshape(nao,-1),
pLqI.reshape(nao,-1), 1,
iLkR.reshape(nao,-1), iLkI.reshape(nao,-1))
iLkR *= vkcoulG[p0:p1].reshape(1,nG,1)
iLkI *= vkcoulG[p0:p1].reshape(1,nG,1)
zdotCN(pLqR.reshape(-1,nao).T, pLqI.reshape(-1,nao).T,
iLkR.reshape(-1,nao), iLkI.reshape(-1,nao),
1, vkR[i,kj], vkI[i,kj], 1)
def _make_j3c(mydf, cell, auxcell, kptij_lst):
t1 = (time.clock(), time.time())
log = logger.Logger(mydf.stdout, mydf.verbose)
max_memory = max(2000, mydf.max_memory-lib.current_memory()[0])
fused_cell, fuse = fuse_auxcell(mydf, mydf.auxcell)
nao = cell.nao_nr()
naux = auxcell.nao_nr()
gs = mydf.gs
Gv, Gvbase, kws = cell.get_Gv_weights(gs)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngs = gxyz.shape[0]
kptis = kptij_lst[:,0]
kptjs = kptij_lst[:,1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('cint2c2e_sph', hermi=1, kpts=uniq_kpts)
kLRs = []
kLIs = []
for k, kpt in enumerate(uniq_kpts):
aoaux = ft_ao.ft_ao(fused_cell, Gv, None, b, gxyz, Gvbase, kpt).T
aoaux = fuse(aoaux)
coulG = numpy.sqrt(mydf.weighted_coulG(kpt, False, gs))
kLR = (aoaux.real * coulG).T
kLI = (aoaux.imag * coulG).T
if not kLR.flags.c_contiguous: kLR = lib.transpose(kLR.T)
if not kLI.flags.c_contiguous: kLI = lib.transpose(kLI.T)
j2c[k] = fuse(fuse(j2c[k]).T).T.copy()
if is_zero(kpt): # kpti == kptj
j2c[k] -= lib.dot(kLR.T, kLR)
j2c[k] -= lib.dot(kLI.T, kLI)
else:
# aoaux ~ kpt_ij, aoaux.conj() ~ kpt_kl
j2cR, j2cI = zdotCN(kLR.T, kLI.T, kLR, kLI)
j2c[k] -= j2cR + j2cI * 1j
w, v = scipy.linalg.eigh(j2c[k])
log.debug('MDF metric for kpt %s cond = %.4g, drop %d bfns',
k, w[0]/w[-1], numpy.count_nonzero(w<df.LINEAR_DEP_THR))
v = v[:,w>df.LINEAR_DEP_THR].T.conj()
v /= numpy.sqrt(w[w>df.LINEAR_DEP_THR]).reshape(-1,1)
j2c[k] = ('eig', v)
kLR *= coulG.reshape(-1,1)
kLI *= coulG.reshape(-1,1)
kLRs.append(kLR)
kLIs.append(kLI)
aoaux = kLR = kLI = j2cR = j2cI = coulG = None
outcore.aux_e2(cell, fused_cell, mydf._cderi, 'cint3c2e_sph',
kptij_lst=kptij_lst, dataname='j3c', max_memory=max_memory)
t1 = log.timer_debug1('3c2e', *t1)
nauxs = [v[1].shape[0] for v in j2c]
feri = h5py.File(mydf._cderi)
def make_kpt(uniq_kptji_id): # kpt = kptj - kpti
kpt = uniq_kpts[uniq_kptji_id]
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
kLR = kLRs[uniq_kptji_id]
kLI = kLIs[uniq_kptji_id]
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao*(nao+1)//2
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('cint1e_ovlp_sph', hermi=1, kpts=adapted_kptjs)
for k, ji in enumerate(adapted_ji_idx):
ovlp[k] = lib.pack_tril(ovlp[k])
else:
aosym = 's1'
nao_pair = nao**2
mem_now = lib.current_memory()[0]
log.debug2('memory = %s', mem_now)
max_memory = max(2000, mydf.max_memory-mem_now)
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(max(int(max_memory*.6*1e6/16/naux/(nkptj+1)), 1), nao_pair)
shranges = _guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(16, int(max_memory*.2*1e6/16/buflen/(nkptj+1)))
else:
Gblksize = max(16, int(max_memory*.4*1e6/16/buflen/(nkptj+1)))
Gblksize = min(Gblksize, ngs, 16384)
pqkRbuf = numpy.empty(buflen*Gblksize)
pqkIbuf = numpy.empty(buflen*Gblksize)
# buf for ft_aopair
buf = numpy.zeros((nkptj,buflen*Gblksize), dtype=numpy.complex128)
col1 = 0
for istep, sh_range in enumerate(shranges):
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', \
istep+1, len(shranges), *sh_range)
bstart, bend, ncol = sh_range
col0, col1 = col1, col1+ncol
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = fuse(numpy.asarray(feri['j3c/%d'%idx][:,col0:col1]))
if is_zero(kpt):
for i, c in enumerate(vbar):
if c != 0:
v[i] -= c * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
v = None
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG,ncol), dtype=numpy.complex128,
order='F', buffer=buf[k])
pqkR = numpy.ndarray((ncol,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
aoao[:] = 0
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k], 1)
else:
shls_slice = (bstart, bend, 0, cell.nbas)
ni = ncol // nao
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG,ni,nao), dtype=numpy.complex128,
order='F', buffer=buf[k])
pqkR = numpy.ndarray((ni,nao,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ni,nao,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.transpose(1,2,0)
pqkI[:] = aoao.imag.transpose(1,2,0)
aoao[:] = 0
pqkR = pqkR.reshape(-1,nG)
pqkI = pqkI.reshape(-1,nG)
zdotCN(kLR[p0:p1].T, kLI[p0:p1].T, pqkR.T, pqkI.T,
-1, j3cR[k], j3cI[k], 1)
naux0 = nauxs[uniq_kptji_id]
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = j3cR[k]
else:
v = j3cR[k] + j3cI[k] * 1j
if j2c[uniq_kptji_id][0] == 'CD':
v = scipy.linalg.solve_triangular(j2c[uniq_kptji_id][1], v,
lower=True, overwrite_b=True)
else:
v = lib.dot(j2c[uniq_kptji_id][1], v)
feri['j3c/%d'%ji][:naux0,col0:col1] = v
naux0 = nauxs[uniq_kptji_id]
for k, ji in enumerate(adapted_ji_idx):
v = feri['j3c/%d'%ji][:naux0]
del(feri['j3c/%d'%ji])
feri['j3c/%d'%ji] = v
for k, kpt in enumerate(uniq_kpts):
make_kpt(k)
feri.close()
def make_kpt(uniq_kptji_id): # kpt = kptj - kpti
kpt = uniq_kpts[uniq_kptji_id]
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
kLR = kLRs[uniq_kptji_id]
kLI = kLIs[uniq_kptji_id]
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao*(nao+1)//2
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('cint1e_ovlp_sph', hermi=1, kpts=adapted_kptjs)
for k, ji in enumerate(adapted_ji_idx):
ovlp[k] = lib.pack_tril(ovlp[k])
else:
aosym = 's1'
nao_pair = nao**2
max_memory = max(2000, mydf.max_memory-lib.current_memory()[0])
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(max(int(max_memory*.6*1e6/16/naux/(nkptj+1)), 1), nao_pair)
shranges = _guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(16, int(max_memory*.2*1e6/16/buflen/(nkptj+1)))
else:
Gblksize = max(16, int(max_memory*.4*1e6/16/buflen/(nkptj+1)))
Gblksize = min(Gblksize, ngs, 16384)
pqkRbuf = numpy.empty(buflen*Gblksize)
pqkIbuf = numpy.empty(buflen*Gblksize)
# buf for ft_aopair
buf = numpy.zeros((nkptj,buflen*Gblksize), dtype=numpy.complex128)
col1 = 0
for istep, sh_range in enumerate(shranges):
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', \
istep+1, len(shranges), *sh_range)
bstart, bend, ncol = sh_range
col0, col1 = col1, col1+ncol
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = numpy.asarray(feri['j3c/%d'%idx][:,col0:col1])
if is_zero(kpt):
for i, c in enumerate(vbar):
if c != 0:
v[i] -= c * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG,ncol), dtype=numpy.complex128,
order='F', buffer=buf[k])
pqkR = numpy.ndarray((ncol,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
aoao[:] = 0
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k][naux:], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k][naux:], 1)
else:
shls_slice = (bstart, bend, 0, cell.nbas)
ni = ncol // nao
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG,ni,nao), dtype=numpy.complex128,
order='F', buffer=buf[k])
pqkR = numpy.ndarray((ni,nao,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ni,nao,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.transpose(1,2,0)
pqkI[:] = aoao.imag.transpose(1,2,0)
aoao[:] = 0
pqkR = pqkR.reshape(-1,nG)
pqkI = pqkI.reshape(-1,nG)
zdotCN(kLR[p0:p1].T, kLI[p0:p1].T, pqkR.T, pqkI.T,
-1, j3cR[k][naux:], j3cI[k][naux:], 1)
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = fuse(j3cR[k])
else:
v = fuse(j3cR[k] + j3cI[k] * 1j)
v = scipy.linalg.solve_triangular(j2c[uniq_kptji_id], v,
lower=True, overwrite_b=True)
feri['j3c/%d'%ji][:naux,col0:col1] = v
def make_kpt(kpt): # kpt = kptj - kpti
# search for all possible ki and kj that has ki-kj+kpt=0
kk_match = numpy.einsum('ijx->ij', abs(kk_table + kpt)) < 1e-9
kpti_idx, kptj_idx = numpy.where(kk_todo & kk_match)
nkptj = len(kptj_idx)
log.debug1('kpt = %s', kpt)
log.debug2('kpti_idx = %s', kpti_idx)
log.debug2('kptj_idx = %s', kptj_idx)
kk_todo[kpti_idx,kptj_idx] = False
if swap_2e and not is_zero(kpt):
kk_todo[kptj_idx,kpti_idx] = False
max_memory1 = max_memory * (nkptj+1)/(nkptj+5)
#blksize = max(int(max_memory1*4e6/(nkptj+5)/16/nao**2), 16)
#bufR = numpy.empty((blksize*nao**2))
#bufI = numpy.empty((blksize*nao**2))
# Use DF object to mimic KRHF/KUHF object in function get_coulG
mydf.exxdiv = exxdiv
vkcoulG = mydf.weighted_coulG(kpt, True, mesh)
kptjs = kpts[kptj_idx]
# <r|-G+k_rs|s> = conj(<s|G-k_rs|r>) = conj(<s|G+k_sr|r>)
#buf1R = numpy.empty((blksize*nao**2))
#buf1I = numpy.empty((blksize*nao**2))
for aoaoks, p0, p1 in mydf.ft_loop(mesh, kpt, kptjs, max_memory=max_memory1):
nG = p1 - p0
bufR = numpy.empty((nG*nao**2))
bufI = numpy.empty((nG*nao**2))
buf1R = numpy.empty((nG*nao**2))
buf1I = numpy.empty((nG*nao**2))
for k, aoao in enumerate(aoaoks):
ki = kpti_idx[k]
kj = kptj_idx[k]
# case 1: k_pq = (pi|iq)
#:v4 = numpy.einsum('ijL,lkL->ijkl', pqk, pqk.conj())
#:vk += numpy.einsum('ijkl,jk->il', v4, dm)
pLqR = numpy.ndarray((nao,nG,nao), buffer=bufR)
pLqI = numpy.ndarray((nao,nG,nao), buffer=bufI)
pLqR[:] = aoao.real.reshape(nG,nao,nao).transpose(1,0,2)
pLqI[:] = aoao.imag.reshape(nG,nao,nao).transpose(1,0,2)
iLkR = numpy.ndarray((nao,nG,nao), buffer=buf1R)
iLkI = numpy.ndarray((nao,nG,nao), buffer=buf1I)
for i in range(nset):
zdotNN(pLqR.reshape(-1,nao), pLqI.reshape(-1,nao),
dmsR[i,kj], dmsI[i,kj], 1,
iLkR.reshape(-1,nao), iLkI.reshape(-1,nao))
iLkR *= vkcoulG[p0:p1].reshape(1,nG,1)
iLkI *= vkcoulG[p0:p1].reshape(1,nG,1)
zdotNC(iLkR.reshape(nao,-1), iLkI.reshape(nao,-1),
pLqR.reshape(nao,-1).T, pLqI.reshape(nao,-1).T,
1, vkR[i,ki], vkI[i,ki], 1)
# case 2: k_pq = (iq|pi)
#:v4 = numpy.einsum('iLj,lLk->ijkl', pqk, pqk.conj())
#:vk += numpy.einsum('ijkl,li->kj', v4, dm)
if swap_2e and not is_zero(kpt):
for i in range(nset):
zdotNN(dmsR[i,ki], dmsI[i,ki], pLqR.reshape(nao,-1),
pLqI.reshape(nao,-1), 1,
iLkR.reshape(nao,-1), iLkI.reshape(nao,-1))
iLkR *= vkcoulG[p0:p1].reshape(1,nG,1)
iLkI *= vkcoulG[p0:p1].reshape(1,nG,1)
zdotCN(pLqR.reshape(-1,nao).T, pLqI.reshape(-1,nao).T,
iLkR.reshape(-1,nao), iLkI.reshape(-1,nao),
1, vkR[i,kj], vkI[i,kj], 1)
def _make_j3c(mydf, cell, auxcell, kptij_lst, cderi_file):
t1 = (time.clock(), time.time())
log = logger.Logger(mydf.stdout, mydf.verbose)
max_memory = max(2000, mydf.max_memory-lib.current_memory()[0])
fused_cell, fuse = fuse_auxcell(mydf, auxcell)
# The ideal way to hold the temporary integrals is to store them in the
# cderi_file and overwrite them inplace in the second pass. The current
# HDF5 library does not have an efficient way to manage free space in
# overwriting. It often leads to the cderi_file ~2 times larger than the
# necessary size. For now, dumping the DF integral intermediates to a
# separated temporary file can avoid this issue. The DF intermediates may
# be terribly huge. The temporary file should be placed in the same disk
# as cderi_file.
swapfile = tempfile.NamedTemporaryFile(dir=os.path.dirname(cderi_file))
fswap = lib.H5TmpFile(swapfile.name)
# Unlink swapfile to avoid trash
swapfile = None
outcore._aux_e2(cell, fused_cell, fswap, 'int3c2e', aosym='s2',
kptij_lst=kptij_lst, dataname='j3c-junk', max_memory=max_memory)
t1 = log.timer_debug1('3c2e', *t1)
nao = cell.nao_nr()
naux = auxcell.nao_nr()
mesh = mydf.mesh
Gv, Gvbase, kws = cell.get_Gv_weights(mesh)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngrids = gxyz.shape[0]
kptis = kptij_lst[:,0]
kptjs = kptij_lst[:,1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
log.debug('Num uniq kpts %d', len(uniq_kpts))
log.debug2('uniq_kpts %s', uniq_kpts)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('int2c2e', hermi=1, kpts=uniq_kpts)
max_memory = max(2000, mydf.max_memory - lib.current_memory()[0])
blksize = max(2048, int(max_memory*.5e6/16/fused_cell.nao_nr()))
log.debug2('max_memory %s (MB) blocksize %s', max_memory, blksize)
for k, kpt in enumerate(uniq_kpts):
coulG = mydf.weighted_coulG(kpt, False, mesh)
for p0, p1 in lib.prange(0, ngrids, blksize):
aoaux = ft_ao.ft_ao(fused_cell, Gv[p0:p1], None, b, gxyz[p0:p1], Gvbase, kpt).T
LkR = numpy.asarray(aoaux.real, order='C')
LkI = numpy.asarray(aoaux.imag, order='C')
aoaux = None
if is_zero(kpt): # kpti == kptj
j2c[k][naux:] -= lib.ddot(LkR[naux:]*coulG[p0:p1], LkR.T)
j2c[k][naux:] -= lib.ddot(LkI[naux:]*coulG[p0:p1], LkI.T)
j2c[k][:naux,naux:] = j2c[k][naux:,:naux].T
else:
j2cR, j2cI = zdotCN(LkR[naux:]*coulG[p0:p1],
LkI[naux:]*coulG[p0:p1], LkR.T, LkI.T)
j2c[k][naux:] -= j2cR + j2cI * 1j
j2c[k][:naux,naux:] = j2c[k][naux:,:naux].T.conj()
LkR = LkI = None
fswap['j2c/%d'%k] = fuse(fuse(j2c[k]).T).T
j2c = coulG = None
def cholesky_decomposed_metric(uniq_kptji_id):
j2c = numpy.asarray(fswap['j2c/%d'%uniq_kptji_id])
j2c_negative = None
try:
j2c = scipy.linalg.cholesky(j2c, lower=True)
j2ctag = 'CD'
except scipy.linalg.LinAlgError as e:
#msg =('===================================\n'
# 'J-metric not positive definite.\n'
# 'It is likely that mesh is not enough.\n'
# '===================================')
#log.error(msg)
#raise scipy.linalg.LinAlgError('\n'.join([str(e), msg]))
w, v = scipy.linalg.eigh(j2c)
log.debug('DF metric linear dependency for kpt %s', uniq_kptji_id)
log.debug('cond = %.4g, drop %d bfns',
w[-1]/w[0], numpy.count_nonzero(w<mydf.linear_dep_threshold))
v1 = v[:,w>mydf.linear_dep_threshold].conj().T
v1 /= numpy.sqrt(w[w>mydf.linear_dep_threshold]).reshape(-1,1)
j2c = v1
if cell.dimension == 2 and cell.low_dim_ft_type != 'inf_vacuum':
idx = numpy.where(w < -mydf.linear_dep_threshold)[0]
if len(idx) > 0:
j2c_negative = (v[:,idx]/numpy.sqrt(-w[idx])).conj().T
w = v = None
j2ctag = 'eig'
return j2c, j2c_negative, j2ctag
feri = h5py.File(cderi_file, 'w')
feri['j3c-kptij'] = kptij_lst
nsegs = len(fswap['j3c-junk/0'])
def make_kpt(uniq_kptji_id, cholesky_j2c):
kpt = uniq_kpts[uniq_kptji_id] # kpt = kptj - kpti
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
j2c, j2c_negative, j2ctag = cholesky_j2c
shls_slice = (auxcell.nbas, fused_cell.nbas)
Gaux = ft_ao.ft_ao(fused_cell, Gv, shls_slice, b, gxyz, Gvbase, kpt)
wcoulG = mydf.weighted_coulG(kpt, False, mesh)
Gaux *= wcoulG.reshape(-1,1)
kLR = Gaux.real.copy('C')
kLI = Gaux.imag.copy('C')
Gaux = None
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao*(nao+1)//2
if cell.dimension == 3:
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('int1e_ovlp', hermi=1, kpts=adapted_kptjs)
ovlp = [lib.pack_tril(s) for s in ovlp]
else:
aosym = 's1'
nao_pair = nao**2
mem_now = lib.current_memory()[0]
log.debug2('memory = %s', mem_now)
max_memory = max(2000, mydf.max_memory-mem_now)
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(max(int(max_memory*.38e6/16/naux/(nkptj+1)), 1), nao_pair)
shranges = _guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(16, int(max_memory*.1e6/16/buflen/(nkptj+1)))
else:
Gblksize = max(16, int(max_memory*.2e6/16/buflen/(nkptj+1)))
Gblksize = min(Gblksize, ngrids, 16384)
pqkRbuf = numpy.empty(buflen*Gblksize)
pqkIbuf = numpy.empty(buflen*Gblksize)
# buf for ft_aopair
buf = numpy.empty(nkptj*buflen*Gblksize, dtype=numpy.complex128)
def pw_contract(istep, sh_range, j3cR, j3cI):
bstart, bend, ncol = sh_range
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
else:
shls_slice = (bstart, bend, 0, cell.nbas)
for p0, p1 in lib.prange(0, ngrids, Gblksize):
dat = ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = dat[k].reshape(nG,ncol)
pqkR = numpy.ndarray((ncol,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k][naux:], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k][naux:], 1)
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = fuse(j3cR[k])
else:
v = fuse(j3cR[k] + j3cI[k] * 1j)
if j2ctag == 'CD':
v = scipy.linalg.solve_triangular(j2c, v, lower=True, overwrite_b=True)
feri['j3c/%d/%d'%(ji,istep)] = v
else:
feri['j3c/%d/%d'%(ji,istep)] = lib.dot(j2c, v)
# low-dimension systems
if j2c_negative is not None:
feri['j3c-/%d/%d'%(ji,istep)] = lib.dot(j2c_negative, v)
with lib.call_in_background(pw_contract) as compute:
col1 = 0
for istep, sh_range in enumerate(shranges):
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', \
istep+1, len(shranges), *sh_range)
bstart, bend, ncol = sh_range
col0, col1 = col1, col1+ncol
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = numpy.vstack([fswap['j3c-junk/%d/%d'%(idx,i)][0,col0:col1].T
for i in range(nsegs)])
# vbar is the interaction between the background charge
# and the auxiliary basis. 0D, 1D, 2D do not have vbar.
if is_zero(kpt) and cell.dimension == 3:
for i in numpy.where(vbar != 0)[0]:
v[i] -= vbar[i] * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
v = None
compute(istep, sh_range, j3cR, j3cI)
for ji in adapted_ji_idx:
del(fswap['j3c-junk/%d'%ji])
# Wrapped around boundary and symmetry between k and -k can be used
# explicitly for the metric integrals. We consider this symmetry
# because it is used in the df_ao2mo module when contracting two 3-index
# integral tensors to the 4-index 2e integral tensor. If the symmetry
# related k-points are treated separately, the resultant 3-index tensors
# may have inconsistent dimension due to the numerial noise when handling
# linear dependency of j2c.
def conj_j2c(cholesky_j2c):
j2c, j2c_negative, j2ctag = cholesky_j2c
if j2c_negative is None:
return j2c.conj(), None, j2ctag
else:
return j2c.conj(), j2c_negative.conj(), j2ctag
a = cell.lattice_vectors() / (2*numpy.pi)
def kconserve_indices(kpt):
'''search which (kpts+kpt) satisfies momentum conservation'''
kdif = numpy.einsum('wx,ix->wi', a, uniq_kpts + kpt)
kdif_int = numpy.rint(kdif)
mask = numpy.einsum('wi->i', abs(kdif - kdif_int)) < KPT_DIFF_TOL
uniq_kptji_ids = numpy.where(mask)[0]
return uniq_kptji_ids
done = numpy.zeros(len(uniq_kpts), dtype=bool)
for k, kpt in enumerate(uniq_kpts):
if done[k]:
continue
log.debug1('Cholesky decomposition for j2c at kpt %s', k)
cholesky_j2c = cholesky_decomposed_metric(k)
# The k-point k' which has (k - k') * a = 2n pi. Metric integrals have the
# symmetry S = S
uniq_kptji_ids = kconserve_indices(-kpt)
log.debug1("Symmetry pattern (k - %s)*a= 2n pi", kpt)
log.debug1(" make_kpt for uniq_kptji_ids %s", uniq_kptji_ids)
for uniq_kptji_id in uniq_kptji_ids:
if not done[uniq_kptji_id]:
make_kpt(uniq_kptji_id, cholesky_j2c)
done[uniq_kptji_ids] = True
# The k-point k' which has (k + k') * a = 2n pi. Metric integrals have the
# symmetry S = S*
uniq_kptji_ids = kconserve_indices(kpt)
log.debug1("Symmetry pattern (k + %s)*a= 2n pi", kpt)
log.debug1(" make_kpt for %s", uniq_kptji_ids)
cholesky_j2c = conj_j2c(cholesky_j2c)
for uniq_kptji_id in uniq_kptji_ids:
if not done[uniq_kptji_id]:
make_kpt(uniq_kptji_id, cholesky_j2c)
done[uniq_kptji_ids] = True
feri.close()
def _make_j3c(mydf, cell, auxcell, kptij_lst, cderi_file):
t1 = (time.clock(), time.time())
log = logger.Logger(mydf.stdout, mydf.verbose)
max_memory = max(2000, mydf.max_memory - lib.current_memory()[0])
fused_cell, fuse = fuse_auxcell(mydf, auxcell)
outcore.aux_e2(cell,
fused_cell,
cderi_file,
'int3c2e_sph',
aosym='s2',
kptij_lst=kptij_lst,
dataname='j3c',
max_memory=max_memory)
t1 = log.timer_debug1('3c2e', *t1)
nao = cell.nao_nr()
naux = auxcell.nao_nr()
gs = mydf.gs
Gv, Gvbase, kws = cell.get_Gv_weights(gs)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngs = gxyz.shape[0]
kptis = kptij_lst[:, 0]
kptjs = kptij_lst[:, 1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
log.debug('Num uniq kpts %d', len(uniq_kpts))
log.debug2('uniq_kpts %s', uniq_kpts)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('int2c2e_sph', hermi=1, kpts=uniq_kpts)
feri = h5py.File(cderi_file)
for k, kpt in enumerate(uniq_kpts):
aoaux = ft_ao.ft_ao(fused_cell, Gv, None, b, gxyz, Gvbase, kpt).T
aoaux = fuse(aoaux)
coulG = numpy.sqrt(mydf.weighted_coulG(kpt, False, gs))
kLR = (aoaux.real * coulG).T
kLI = (aoaux.imag * coulG).T
if not kLR.flags.c_contiguous: kLR = lib.transpose(kLR.T)
if not kLI.flags.c_contiguous: kLI = lib.transpose(kLI.T)
j2c_k = fuse(fuse(j2c[k]).T).T.copy()
if is_zero(kpt): # kpti == kptj
j2c_k -= lib.dot(kLR.T, kLR)
j2c_k -= lib.dot(kLI.T, kLI)
else:
# aoaux ~ kpt_ij, aoaux.conj() ~ kpt_kl
j2cR, j2cI = zdotCN(kLR.T, kLI.T, kLR, kLI)
j2c_k -= j2cR + j2cI * 1j
feri['j2c/%d' % k] = j2c_k
aoaux = kLR = kLI = j2cR = j2cI = coulG = None
j2c = None
def make_kpt(uniq_kptji_id): # kpt = kptj - kpti
kpt = uniq_kpts[uniq_kptji_id]
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
Gaux = ft_ao.ft_ao(fused_cell, Gv, None, b, gxyz, Gvbase, kpt).T
Gaux = fuse(Gaux)
Gaux *= mydf.weighted_coulG(kpt, False, gs)
kLR = Gaux.T.real.copy('C')
kLI = Gaux.T.imag.copy('C')
j2c = numpy.asarray(feri['j2c/%d' % uniq_kptji_id])
# Note large difference may be found in results between the CD/eig treatments.
# In some systems, small integral errors can lead to different treatments of
# linear dependency which can be observed in the total energy/orbital energy
# around 4th decimal place.
# try:
# j2c = scipy.linalg.cholesky(j2c, lower=True)
# j2ctag = 'CD'
# except scipy.linalg.LinAlgError as e:
#
# Abandon CD treatment for better numerical stablity
w, v = scipy.linalg.eigh(j2c)
log.debug('MDF metric for kpt %s cond = %.4g, drop %d bfns',
uniq_kptji_id, w[-1] / w[0],
numpy.count_nonzero(w < mydf.linear_dep_threshold))
v = v[:, w > mydf.linear_dep_threshold].T.conj()
v /= numpy.sqrt(w[w > mydf.linear_dep_threshold]).reshape(-1, 1)
j2c = v
j2ctag = 'eig'
naux0 = j2c.shape[0]
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao * (nao + 1) // 2
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('int1e_ovlp_sph',
hermi=1,
kpts=adapted_kptjs)
for k, ji in enumerate(adapted_ji_idx):
ovlp[k] = lib.pack_tril(ovlp[k])
else:
aosym = 's1'
nao_pair = nao**2
mem_now = lib.current_memory()[0]
log.debug2('memory = %s', mem_now)
max_memory = max(2000, mydf.max_memory - mem_now)
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(
max(int(max_memory * .6 * 1e6 / 16 / naux / (nkptj + 1)), 1),
nao_pair)
shranges = _guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(
16, int(max_memory * .2 * 1e6 / 16 / buflen / (nkptj + 1)))
else:
Gblksize = max(
16, int(max_memory * .4 * 1e6 / 16 / buflen / (nkptj + 1)))
Gblksize = min(Gblksize, ngs, 16384)
pqkRbuf = numpy.empty(buflen * Gblksize)
pqkIbuf = numpy.empty(buflen * Gblksize)
# buf for ft_aopair
buf = numpy.empty((nkptj, buflen * Gblksize), dtype=numpy.complex128)
col1 = 0
for istep, sh_range in enumerate(shranges):
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', \
istep+1, len(shranges), *sh_range)
bstart, bend, ncol = sh_range
col0, col1 = col1, col1 + ncol
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = fuse(numpy.asarray(feri['j3c/%d' % idx][:, col0:col1]))
if is_zero(kpt):
for i, c in enumerate(vbar):
if c != 0:
v[i] -= c * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
v = None
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
else:
shls_slice = (bstart, bend, 0, cell.nbas)
for p0, p1 in lib.prange(0, ngs, Gblksize):
dat = ft_ao._ft_aopair_kpts(cell,
Gv[p0:p1],
shls_slice,
aosym,
b,
gxyz[p0:p1],
Gvbase,
kpt,
adapted_kptjs,
out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = dat[k].reshape(nG, ncol)
pqkR = numpy.ndarray((ncol, nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol, nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k], 1)
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = j3cR[k]
else:
v = j3cR[k] + j3cI[k] * 1j
if j2ctag == 'CD':
v = scipy.linalg.solve_triangular(j2c,
v,
lower=True,
overwrite_b=True)
else:
v = lib.dot(j2c, v)
feri['j3c/%d' % ji][:naux0, col0:col1] = v
del (feri['j2c/%d' % uniq_kptji_id])
for k, ji in enumerate(adapted_ji_idx):
v = feri['j3c/%d' % ji][:naux0]
del (feri['j3c/%d' % ji])
feri['j3c/%d' % ji] = v
for k, kpt in enumerate(uniq_kpts):
make_kpt(k)
feri.close()
def _make_j3c(mydf, cell, auxcell, kptij_lst, cderi_file):
t1 = (time.clock(), time.time())
log = logger.Logger(mydf.stdout, mydf.verbose)
max_memory = max(2000, mydf.max_memory-lib.current_memory()[0])
fused_cell, fuse = fuse_auxcell(mydf, auxcell)
# The ideal way to hold the temporary integrals is to store them in the
# cderi_file and overwrite them inplace in the second pass. The current
# HDF5 library does not have an efficient way to manage free space in
# overwriting. It often leads to the cderi_file ~2 times larger than the
# necessary size. For now, dumping the DF integral intermediates to a
# separated temporary file can avoid this issue. The DF intermediates may
# be terribly huge. The temporary file should be placed in the same disk
# as cderi_file.
swapfile = tempfile.NamedTemporaryFile(dir=os.path.dirname(cderi_file))
fswap = lib.H5TmpFile(swapfile.name)
# Unlink swapfile to avoid trash
swapfile = None
outcore._aux_e2(cell, fused_cell, fswap, 'int3c2e', aosym='s2',
kptij_lst=kptij_lst, dataname='j3c-junk', max_memory=max_memory)
t1 = log.timer_debug1('3c2e', *t1)
nao = cell.nao_nr()
naux = auxcell.nao_nr()
mesh = mydf.mesh
Gv, Gvbase, kws = cell.get_Gv_weights(mesh)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngrids = gxyz.shape[0]
kptis = kptij_lst[:,0]
kptjs = kptij_lst[:,1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
log.debug('Num uniq kpts %d', len(uniq_kpts))
log.debug2('uniq_kpts %s', uniq_kpts)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('int2c2e', hermi=1, kpts=uniq_kpts)
# An alternative method to evalute j2c. This method might have larger numerical error?
# chgcell = make_modchg_basis(auxcell, mydf.eta)
# for k, kpt in enumerate(uniq_kpts):
# aoaux = ft_ao.ft_ao(chgcell, Gv, None, b, gxyz, Gvbase, kpt).T
# coulG = numpy.sqrt(mydf.weighted_coulG(kpt, False, mesh))
# LkR = aoaux.real * coulG
# LkI = aoaux.imag * coulG
# j2caux = numpy.zeros_like(j2c[k])
# j2caux[naux:,naux:] = j2c[k][naux:,naux:]
# if is_zero(kpt): # kpti == kptj
# j2caux[naux:,naux:] -= lib.ddot(LkR, LkR.T)
# j2caux[naux:,naux:] -= lib.ddot(LkI, LkI.T)
# j2c[k] = j2c[k][:naux,:naux] - fuse(fuse(j2caux.T).T)
# vbar = fuse(mydf.auxbar(fused_cell))
# s = (vbar != 0).astype(numpy.double)
# j2c[k] -= numpy.einsum('i,j->ij', vbar, s)
# j2c[k] -= numpy.einsum('i,j->ij', s, vbar)
# else:
# j2cR, j2cI = zdotCN(LkR, LkI, LkR.T, LkI.T)
# j2caux[naux:,naux:] -= j2cR + j2cI * 1j
# j2c[k] = j2c[k][:naux,:naux] - fuse(fuse(j2caux.T).T)
# fswap['j2c/%d'%k] = fuse(fuse(j2c[k]).T).T
# aoaux = LkR = LkI = coulG = None
if cell.dimension == 1 or cell.dimension == 2:
plain_ints = _gaussian_int(fused_cell)
max_memory = max(2000, mydf.max_memory - lib.current_memory()[0])
blksize = max(2048, int(max_memory*.5e6/16/fused_cell.nao_nr()))
log.debug2('max_memory %s (MB) blocksize %s', max_memory, blksize)
for k, kpt in enumerate(uniq_kpts):
coulG = numpy.sqrt(mydf.weighted_coulG(kpt, False, mesh))
for p0, p1 in lib.prange(0, ngrids, blksize):
aoaux = ft_ao.ft_ao(fused_cell, Gv[p0:p1], None, b, gxyz[p0:p1], Gvbase, kpt)
if (cell.dimension == 1 or cell.dimension == 2) and is_zero(kpt):
G0idx, SI_on_z = pbcgto.cell._SI_for_uniform_model_charge(cell, Gv[p0:p1])
aoaux[G0idx] -= numpy.einsum('g,i->gi', SI_on_z, plain_ints)
aoaux = aoaux.T
LkR = aoaux.real * coulG[p0:p1]
LkI = aoaux.imag * coulG[p0:p1]
aoaux = None
if is_zero(kpt): # kpti == kptj
j2c[k][naux:] -= lib.ddot(LkR[naux:], LkR.T)
j2c[k][naux:] -= lib.ddot(LkI[naux:], LkI.T)
j2c[k][:naux,naux:] = j2c[k][naux:,:naux].T
else:
j2cR, j2cI = zdotCN(LkR[naux:], LkI[naux:], LkR.T, LkI.T)
j2c[k][naux:] -= j2cR + j2cI * 1j
j2c[k][:naux,naux:] = j2c[k][naux:,:naux].T.conj()
LkR = LkI = None
fswap['j2c/%d'%k] = fuse(fuse(j2c[k]).T).T
j2c = coulG = None
feri = h5py.File(cderi_file, 'w')
feri['j3c-kptij'] = kptij_lst
nsegs = len(fswap['j3c-junk/0'])
def make_kpt(uniq_kptji_id): # kpt = kptj - kpti
kpt = uniq_kpts[uniq_kptji_id]
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
shls_slice = (auxcell.nbas, fused_cell.nbas)
Gaux = ft_ao.ft_ao(fused_cell, Gv, shls_slice, b, gxyz, Gvbase, kpt)
if (cell.dimension == 1 or cell.dimension == 2) and is_zero(kpt):
G0idx, SI_on_z = pbcgto.cell._SI_for_uniform_model_charge(cell, Gv)
s = plain_ints[-Gaux.shape[1]:] # Only compensated Gaussians
Gaux[G0idx] -= numpy.einsum('g,i->gi', SI_on_z, s)
wcoulG = mydf.weighted_coulG(kpt, False, mesh)
Gaux *= wcoulG.reshape(-1,1)
kLR = Gaux.real.copy('C')
kLI = Gaux.imag.copy('C')
Gaux = None
j2c = numpy.asarray(fswap['j2c/%d'%uniq_kptji_id])
try:
j2c = scipy.linalg.cholesky(j2c, lower=True)
j2ctag = 'CD'
except scipy.linalg.LinAlgError as e:
#msg =('===================================\n'
# 'J-metric not positive definite.\n'
# 'It is likely that mesh is not enough.\n'
# '===================================')
#log.error(msg)
#raise scipy.linalg.LinAlgError('\n'.join([e.message, msg]))
w, v = scipy.linalg.eigh(j2c)
log.debug('DF metric linear dependency for kpt %s', uniq_kptji_id)
log.debug('cond = %.4g, drop %d bfns',
w[-1]/w[0], numpy.count_nonzero(w<mydf.linear_dep_threshold))
v = v[:,w>mydf.linear_dep_threshold].T.conj()
v /= numpy.sqrt(w[w>mydf.linear_dep_threshold]).reshape(-1,1)
j2c = v
j2ctag = 'eig'
naux0 = j2c.shape[0]
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao*(nao+1)//2
vbar = mydf.auxbar(fused_cell)
ovlp = cell.pbc_intor('int1e_ovlp', hermi=1, kpts=adapted_kptjs)
ovlp = [lib.pack_tril(s) for s in ovlp]
else:
aosym = 's1'
nao_pair = nao**2
mem_now = lib.current_memory()[0]
log.debug2('memory = %s', mem_now)
max_memory = max(2000, mydf.max_memory-mem_now)
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(max(int(max_memory*.38e6/16/naux/(nkptj+1)), 1), nao_pair)
shranges = _guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(16, int(max_memory*.1e6/16/buflen/(nkptj+1)))
else:
Gblksize = max(16, int(max_memory*.2e6/16/buflen/(nkptj+1)))
Gblksize = min(Gblksize, ngrids, 16384)
pqkRbuf = numpy.empty(buflen*Gblksize)
pqkIbuf = numpy.empty(buflen*Gblksize)
# buf for ft_aopair
buf = numpy.empty(nkptj*buflen*Gblksize, dtype=numpy.complex128)
def pw_contract(istep, sh_range, j3cR, j3cI):
bstart, bend, ncol = sh_range
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
else:
shls_slice = (bstart, bend, 0, cell.nbas)
for p0, p1 in lib.prange(0, ngrids, Gblksize):
dat = ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
if (cell.dimension == 1 or cell.dimension == 2) and is_zero(kpt):
G0idx, SI_on_z = pbcgto.cell._SI_for_uniform_model_charge(cell, Gv[p0:p1])
if SI_on_z.size > 0:
for k, aoao in enumerate(dat):
aoao[G0idx] -= numpy.einsum('g,i->gi', SI_on_z, ovlp[k])
aux = fuse(ft_ao.ft_ao(fused_cell, Gv[p0:p1][G0idx]).T)
vG_mod = numpy.einsum('ig,g,g->i', aux.conj(),
wcoulG[p0:p1][G0idx], SI_on_z)
if gamma_point(adapted_kptjs[k]):
j3cR[k][:naux] -= vG_mod[:,None].real * ovlp[k]
else:
tmp = vG_mod[:,None] * ovlp[k]
j3cR[k][:naux] -= tmp.real
j3cI[k][:naux] -= tmp.imag
tmp = aux = vG_mod
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = dat[k].reshape(nG,ncol)
pqkR = numpy.ndarray((ncol,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k][naux:], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k][naux:], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k][naux:], 1)
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = fuse(j3cR[k])
else:
v = fuse(j3cR[k] + j3cI[k] * 1j)
if j2ctag == 'CD':
v = scipy.linalg.solve_triangular(j2c, v, lower=True, overwrite_b=True)
else:
v = lib.dot(j2c, v)
feri['j3c/%d/%d'%(ji,istep)] = v
with lib.call_in_background(pw_contract) as compute:
col1 = 0
for istep, sh_range in enumerate(shranges):
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', \
istep+1, len(shranges), *sh_range)
bstart, bend, ncol = sh_range
col0, col1 = col1, col1+ncol
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = numpy.vstack([fswap['j3c-junk/%d/%d'%(idx,i)][0,col0:col1].T
for i in range(nsegs)])
if is_zero(kpt) and cell.dimension == 3:
for i in numpy.where(vbar != 0)[0]:
v[i] -= vbar[i] * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
v = None
compute(istep, sh_range, j3cR, j3cI)
for ji in adapted_ji_idx:
del(fswap['j3c-junk/%d'%ji])
for k, kpt in enumerate(uniq_kpts):
make_kpt(k)
feri.close()
def _make_j3c(mydf, cell, auxcell, kptij_lst, cderi_file):
log = logger.Logger(mydf.stdout, mydf.verbose)
t1 = t0 = (time.clock(), time.time())
fused_cell, fuse = fuse_auxcell(mydf, mydf.auxcell)
ao_loc = cell.ao_loc_nr()
nao = ao_loc[-1]
naux = auxcell.nao_nr()
nkptij = len(kptij_lst)
mesh = mydf.mesh
Gv, Gvbase, kws = cell.get_Gv_weights(mesh)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngrids = gxyz.shape[0]
kptis = kptij_lst[:, 0]
kptjs = kptij_lst[:, 1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
log.debug('Num uniq kpts %d', len(uniq_kpts))
log.debug2('uniq_kpts %s', uniq_kpts)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('int2c2e', hermi=1, kpts=uniq_kpts)
j2ctags = []
t1 = log.timer_debug1('2c2e', *t1)
swapfile = tempfile.NamedTemporaryFile(dir=os.path.dirname(cderi_file))
fswap = lib.H5TmpFile(swapfile.name)
# Unlink swapfile to avoid trash
swapfile = None
mem_now = max(comm.allgather(lib.current_memory()[0]))
max_memory = max(2000, mydf.max_memory - mem_now)
blksize = max(2048, int(max_memory * .5e6 / 16 / fused_cell.nao_nr()))
log.debug2('max_memory %s (MB) blocksize %s', max_memory, blksize)
for k, kpt in enumerate(uniq_kpts):
coulG = mydf.weighted_coulG(kpt, False, mesh)
j2c_k = numpy.zeros_like(j2c[k])
for p0, p1 in mydf.prange(0, ngrids, blksize):
aoaux = ft_ao.ft_ao(fused_cell, Gv[p0:p1], None, b, gxyz[p0:p1],
Gvbase, kpt).T
LkR = numpy.asarray(aoaux.real, order='C')
LkI = numpy.asarray(aoaux.imag, order='C')
aoaux = None
if is_zero(kpt): # kpti == kptj
j2c_k[naux:] += lib.ddot(LkR[naux:] * coulG[p0:p1], LkR.T)
j2c_k[naux:] += lib.ddot(LkI[naux:] * coulG[p0:p1], LkI.T)
else:
j2cR, j2cI = zdotCN(LkR[naux:] * coulG[p0:p1],
LkI[naux:] * coulG[p0:p1], LkR.T, LkI.T)
j2c_k[naux:] += j2cR + j2cI * 1j
kLR = kLI = None
j2c_k[:naux, naux:] = j2c_k[naux:, :naux].conj().T
j2c[k] -= mpi.allreduce(j2c_k)
j2c[k] = fuse(fuse(j2c[k]).T).T
try:
fswap['j2c/%d' % k] = scipy.linalg.cholesky(j2c[k], lower=True)
j2ctags.append('CD')
except scipy.linalg.LinAlgError as e:
#msg =('===================================\n'
# 'J-metric not positive definite.\n'
# 'It is likely that mesh is not enough.\n'
# '===================================')
#log.error(msg)
#raise scipy.linalg.LinAlgError('\n'.join([str(e), msg]))
w, v = scipy.linalg.eigh(j2c[k])
log.debug2('metric linear dependency for kpt %s', k)
log.debug2('cond = %.4g, drop %d bfns', w[0] / w[-1],
numpy.count_nonzero(w < mydf.linear_dep_threshold))
v1 = v[:, w > mydf.linear_dep_threshold].T.conj()
v1 /= numpy.sqrt(w[w > mydf.linear_dep_threshold]).reshape(-1, 1)
fswap['j2c/%d' % k] = v1
if cell.dimension == 2 and cell.low_dim_ft_type != 'inf_vacuum':
idx = numpy.where(w < -mydf.linear_dep_threshold)[0]
if len(idx) > 0:
fswap['j2c-/%d' % k] = (v[:, idx] /
numpy.sqrt(-w[idx])).conj().T
w = v = v1 = None
j2ctags.append('eig')
j2c = coulG = None
aosym_s2 = numpy.einsum('ix->i', abs(kptis - kptjs)) < 1e-9
j_only = numpy.all(aosym_s2)
if gamma_point(kptij_lst):
dtype = 'f8'
else:
dtype = 'c16'
t1 = log.timer_debug1('aoaux and int2c', *t1)
# Estimates the buffer size based on the last contraction in G-space.
# This contraction requires to hold nkptj copies of (naux,?) array
# simultaneously in memory.
mem_now = max(comm.allgather(lib.current_memory()[0]))
max_memory = max(2000, mydf.max_memory - mem_now)
nkptj_max = max((uniq_inverse == x).sum() for x in set(uniq_inverse))
buflen = max(
int(
min(max_memory * .5e6 / 16 / naux / (nkptj_max + 2) / nao,
nao / 3 / mpi.pool.size)), 1)
chunks = (buflen, nao)
j3c_jobs = grids2d_int3c_jobs(cell, auxcell, kptij_lst, chunks, j_only)
log.debug1('max_memory = %d MB (%d in use) chunks %s', max_memory,
mem_now, chunks)
log.debug2('j3c_jobs %s', j3c_jobs)
if j_only:
int3c = wrap_int3c(cell, fused_cell, 'int3c2e', 's2', 1, kptij_lst)
else:
int3c = wrap_int3c(cell, fused_cell, 'int3c2e', 's1', 1, kptij_lst)
idxb = numpy.tril_indices(nao)
idxb = (idxb[0] * nao + idxb[1]).astype('i')
aux_loc = fused_cell.ao_loc_nr('ssc' in 'int3c2e')
def gen_int3c(job_id, ish0, ish1):
dataname = 'j3c-chunks/%d' % job_id
i0 = ao_loc[ish0]
i1 = ao_loc[ish1]
dii = i1 * (i1 + 1) // 2 - i0 * (i0 + 1) // 2
if j_only:
dij = dii
buflen = max(8, int(max_memory * 1e6 / 16 / (nkptij * dii + dii)))
else:
dij = (i1 - i0) * nao
buflen = max(8, int(max_memory * 1e6 / 16 / (nkptij * dij + dij)))
auxranges = balance_segs(aux_loc[1:] - aux_loc[:-1], buflen)
buflen = max([x[2] for x in auxranges])
buf = numpy.empty(nkptij * dij * buflen, dtype=dtype)
buf1 = numpy.empty(dij * buflen, dtype=dtype)
naux = aux_loc[-1]
for kpt_id, kptij in enumerate(kptij_lst):
key = '%s/%d' % (dataname, kpt_id)
if aosym_s2[kpt_id]:
shape = (naux, dii)
else:
shape = (naux, dij)
if gamma_point(kptij):
fswap.create_dataset(key, shape, 'f8')
else:
fswap.create_dataset(key, shape, 'c16')
naux0 = 0
for istep, auxrange in enumerate(auxranges):
log.alldebug2("aux_e1 job_id %d step %d", job_id, istep)
sh0, sh1, nrow = auxrange
sub_slice = (ish0, ish1, 0, cell.nbas, sh0, sh1)
mat = numpy.ndarray((nkptij, dij, nrow), dtype=dtype, buffer=buf)
mat = int3c(sub_slice, mat)
for k, kptij in enumerate(kptij_lst):
h5dat = fswap['%s/%d' % (dataname, k)]
v = lib.transpose(mat[k], out=buf1)
if not j_only and aosym_s2[k]:
idy = idxb[i0 * (i0 + 1) // 2:i1 *
(i1 + 1) // 2] - i0 * nao
out = numpy.ndarray((nrow, dii),
dtype=v.dtype,
buffer=mat[k])
v = numpy.take(v, idy, axis=1, out=out)
if gamma_point(kptij):
h5dat[naux0:naux0 + nrow] = v.real
else:
h5dat[naux0:naux0 + nrow] = v
naux0 += nrow
def ft_fuse(job_id, uniq_kptji_id, sh0, sh1):
kpt = uniq_kpts[uniq_kptji_id] # kpt = kptj - kpti
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
j2c = numpy.asarray(fswap['j2c/%d' % uniq_kptji_id])
j2ctag = j2ctags[uniq_kptji_id]
naux0 = j2c.shape[0]
if ('j2c-/%d' % uniq_kptji_id) in fswap:
j2c_negative = numpy.asarray(fswap['j2c-/%d' % uniq_kptji_id])
else:
j2c_negative = None
if is_zero(kpt):
aosym = 's2'
else:
aosym = 's1'
if aosym == 's2' and cell.dimension == 3:
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('int1e_ovlp', hermi=1, kpts=adapted_kptjs)
ovlp = [lib.pack_tril(s) for s in ovlp]
j3cR = [None] * nkptj
j3cI = [None] * nkptj
i0 = ao_loc[sh0]
i1 = ao_loc[sh1]
for k, idx in enumerate(adapted_ji_idx):
key = 'j3c-chunks/%d/%d' % (job_id, idx)
v = numpy.asarray(fswap[key])
if aosym == 's2' and cell.dimension == 3:
for i in numpy.where(vbar != 0)[0]:
v[i] -= vbar[i] * ovlp[k][i0 * (i0 + 1) // 2:i1 *
(i1 + 1) // 2].ravel()
j3cR[k] = numpy.asarray(v.real, order='C')
if v.dtype == numpy.complex128:
j3cI[k] = numpy.asarray(v.imag, order='C')
v = None
ncol = j3cR[0].shape[1]
Gblksize = max(16, int(max_memory * 1e6 / 16 / ncol /
(nkptj + 1))) # +1 for pqkRbuf/pqkIbuf
Gblksize = min(Gblksize, ngrids, 16384)
pqkRbuf = numpy.empty(ncol * Gblksize)
pqkIbuf = numpy.empty(ncol * Gblksize)
buf = numpy.empty(nkptj * ncol * Gblksize, dtype=numpy.complex128)
log.alldebug2('job_id %d blksize (%d,%d)', job_id, Gblksize, ncol)
wcoulG = mydf.weighted_coulG(kpt, False, mesh)
fused_cell_slice = (auxcell.nbas, fused_cell.nbas)
if aosym == 's2':
shls_slice = (sh0, sh1, 0, sh1)
else:
shls_slice = (sh0, sh1, 0, cell.nbas)
for p0, p1 in lib.prange(0, ngrids, Gblksize):
Gaux = ft_ao.ft_ao(fused_cell, Gv[p0:p1], fused_cell_slice, b,
gxyz[p0:p1], Gvbase, kpt)
Gaux *= wcoulG[p0:p1, None]
kLR = Gaux.real.copy('C')
kLI = Gaux.imag.copy('C')
Gaux = None
dat = ft_ao._ft_aopair_kpts(cell,
Gv[p0:p1],
shls_slice,
aosym,
b,
gxyz[p0:p1],
Gvbase,
kpt,
adapted_kptjs,
out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = dat[k].reshape(nG, ncol)
pqkR = numpy.ndarray((ncol, nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol, nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
lib.dot(kLR.T, pqkR.T, -1, j3cR[k][naux:], 1)
lib.dot(kLI.T, pqkI.T, -1, j3cR[k][naux:], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR.T, pqkI.T, -1, j3cI[k][naux:], 1)
lib.dot(kLI.T, pqkR.T, 1, j3cI[k][naux:], 1)
kLR = kLI = None
for k, idx in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
v = fuse(j3cR[k])
else:
v = fuse(j3cR[k] + j3cI[k] * 1j)
if j2ctag == 'CD':
v = scipy.linalg.solve_triangular(j2c,
v,
lower=True,
overwrite_b=True)
fswap['j3c-chunks/%d/%d' % (job_id, idx)][:naux0] = v
else:
fswap['j3c-chunks/%d/%d' % (job_id, idx)][:naux0] = lib.dot(
j2c, v)
# low-dimension systems
if j2c_negative is not None:
fswap['j3c-/%d/%d' % (job_id, idx)] = lib.dot(j2c_negative, v)
_assemble(mydf, kptij_lst, j3c_jobs, gen_int3c, ft_fuse, cderi_file, fswap,
log)
def _make_j3c(mydf, cell, auxcell, kptij_lst):
t1 = (time.clock(), time.time())
log = logger.Logger(mydf.stdout, mydf.verbose)
max_memory = max(2000, mydf.max_memory-lib.current_memory()[0])
fused_cell, fuse = fuse_auxcell(mydf, mydf.auxcell)
if mydf.metric.upper() != 'J':
outcore.aux_e2(cell, fused_cell, mydf._cderi, 'cint3c2e_sph',
kptij_lst=kptij_lst, dataname='j3c', max_memory=max_memory)
t1 = log.timer_debug1('3c2e', *t1)
nao = cell.nao_nr()
naux = auxcell.nao_nr()
gs = mydf.gs
Gv, Gvbase, kws = cell.get_Gv_weights(gs)
b = cell.reciprocal_vectors()
gxyz = lib.cartesian_prod([numpy.arange(len(x)) for x in Gvbase])
ngs = gxyz.shape[0]
kptis = kptij_lst[:,0]
kptjs = kptij_lst[:,1]
kpt_ji = kptjs - kptis
uniq_kpts, uniq_index, uniq_inverse = unique(kpt_ji)
# j2c ~ (-kpt_ji | kpt_ji)
j2c = fused_cell.pbc_intor('cint2c2e_sph', hermi=1, kpts=uniq_kpts)
kLRs = []
kLIs = []
for k, kpt in enumerate(uniq_kpts):
aoaux = ft_ao.ft_ao(fused_cell, Gv, None, b, gxyz, Gvbase, kpt).T
aoaux = fuse(aoaux)
coulG = numpy.sqrt(mydf.weighted_coulG(kpt, False, gs))
kLR = (aoaux.real * coulG).T
kLI = (aoaux.imag * coulG).T
if not kLR.flags.c_contiguous: kLR = lib.transpose(kLR.T)
if not kLI.flags.c_contiguous: kLI = lib.transpose(kLI.T)
j2c[k] = fuse(fuse(j2c[k]).T).T.copy()
if is_zero(kpt): # kpti == kptj
j2c[k] -= lib.dot(kLR.T, kLR)
j2c[k] -= lib.dot(kLI.T, kLI)
else:
# aoaux ~ kpt_ij, aoaux.conj() ~ kpt_kl
j2cR, j2cI = zdotCN(kLR.T, kLI.T, kLR, kLI)
j2c[k] -= j2cR + j2cI * 1j
kLR *= coulG.reshape(-1,1)
kLI *= coulG.reshape(-1,1)
kLRs.append(kLR)
kLIs.append(kLI)
aoaux = kLR = kLI = j2cR = j2cI = coulG = None
feri = h5py.File(mydf._cderi)
log.debug2('memory = %s', lib.current_memory()[0])
# Expand approx Lpq for aosym='s1'. The approx Lpq are all in aosym='s2' mode
if mydf.approx_sr_level > 0 and len(kptij_lst) > 1:
Lpq_fake = _fake_Lpq_kpts(mydf, feri, naux, nao)
def save(label, dat, col0, col1):
nrow = dat.shape[0]
feri[label][:nrow,col0:col1] = dat
def make_kpt(uniq_kptji_id): # kpt = kptj - kpti
kpt = uniq_kpts[uniq_kptji_id]
log.debug1('kpt = %s', kpt)
adapted_ji_idx = numpy.where(uniq_inverse == uniq_kptji_id)[0]
adapted_kptjs = kptjs[adapted_ji_idx]
nkptj = len(adapted_kptjs)
log.debug1('adapted_ji_idx = %s', adapted_ji_idx)
kLR = kLRs[uniq_kptji_id]
kLI = kLIs[uniq_kptji_id]
if is_zero(kpt): # kpti == kptj
aosym = 's2'
nao_pair = nao*(nao+1)//2
vbar = fuse(mydf.auxbar(fused_cell))
ovlp = cell.pbc_intor('cint1e_ovlp_sph', hermi=1, kpts=adapted_kptjs)
for k, ji in enumerate(adapted_ji_idx):
ovlp[k] = lib.pack_tril(ovlp[k])
else:
aosym = 's1'
nao_pair = nao**2
mem_now = lib.current_memory()[0]
log.debug2('memory = %s', mem_now)
max_memory = max(2000, mydf.max_memory-mem_now)
# nkptj for 3c-coulomb arrays plus 1 Lpq array
buflen = min(max(int(max_memory*.6*1e6/16/naux/(nkptj+1)), 1), nao_pair)
shranges = pyscf.df.outcore._guess_shell_ranges(cell, buflen, aosym)
buflen = max([x[2] for x in shranges])
# +1 for a pqkbuf
if aosym == 's2':
Gblksize = max(16, int(max_memory*.2*1e6/16/buflen/(nkptj+1)))
else:
Gblksize = max(16, int(max_memory*.4*1e6/16/buflen/(nkptj+1)))
Gblksize = min(Gblksize, ngs, 16384)
pqkRbuf = numpy.empty(buflen*Gblksize)
pqkIbuf = numpy.empty(buflen*Gblksize)
# buf for ft_aopair
buf = numpy.zeros((nkptj,buflen*Gblksize), dtype=numpy.complex128)
col1 = 0
for istep, sh_range in enumerate(shranges):
log.debug1('int3c2e [%d/%d], AO [%d:%d], ncol = %d', \
istep+1, len(shranges), *sh_range)
bstart, bend, ncol = sh_range
col0, col1 = col1, col1+ncol
j3cR = []
j3cI = []
for k, idx in enumerate(adapted_ji_idx):
v = fuse(numpy.asarray(feri['j3c/%d'%idx][:,col0:col1]))
if mydf.approx_sr_level == 0:
Lpq = numpy.asarray(feri['Lpq/%d'%idx][:,col0:col1])
elif aosym == 's2':
Lpq = numpy.asarray(feri['Lpq/0'][:,col0:col1])
else:
Lpq = numpy.asarray(Lpq_fake[:,col0:col1])
lib.dot(j2c[uniq_kptji_id], Lpq, -.5, v, 1)
if is_zero(kpt):
for i, c in enumerate(vbar):
if c != 0:
v[i] -= c * ovlp[k][col0:col1]
j3cR.append(numpy.asarray(v.real, order='C'))
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
j3cI.append(None)
else:
j3cI.append(numpy.asarray(v.imag, order='C'))
v = Lpq = None
log.debug3(' istep, k = %d %d memory = %s',
istep, k, lib.current_memory()[0])
if aosym == 's2':
shls_slice = (bstart, bend, 0, bend)
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG,ncol), dtype=numpy.complex128,
order='F', buffer=buf[k])
pqkR = numpy.ndarray((ncol,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ncol,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.T
pqkI[:] = aoao.imag.T
aoao[:] = 0
lib.dot(kLR[p0:p1].T, pqkR.T, -1, j3cR[k], 1)
lib.dot(kLI[p0:p1].T, pqkI.T, -1, j3cR[k], 1)
if not (is_zero(kpt) and gamma_point(adapted_kptjs[k])):
lib.dot(kLR[p0:p1].T, pqkI.T, -1, j3cI[k], 1)
lib.dot(kLI[p0:p1].T, pqkR.T, 1, j3cI[k], 1)
log.debug3(' p0:p1 = %d:%d memory = %s',
p0, p1, lib.current_memory()[0])
else:
shls_slice = (bstart, bend, 0, cell.nbas)
ni = ncol // nao
for p0, p1 in lib.prange(0, ngs, Gblksize):
ft_ao._ft_aopair_kpts(cell, Gv[p0:p1], shls_slice, aosym,
b, gxyz[p0:p1], Gvbase, kpt,
adapted_kptjs, out=buf)
nG = p1 - p0
for k, ji in enumerate(adapted_ji_idx):
aoao = numpy.ndarray((nG,ni,nao), dtype=numpy.complex128,
order='F', buffer=buf[k])
pqkR = numpy.ndarray((ni,nao,nG), buffer=pqkRbuf)
pqkI = numpy.ndarray((ni,nao,nG), buffer=pqkIbuf)
pqkR[:] = aoao.real.transpose(1,2,0)
pqkI[:] = aoao.imag.transpose(1,2,0)
aoao[:] = 0
pqkR = pqkR.reshape(-1,nG)
pqkI = pqkI.reshape(-1,nG)
zdotCN(kLR[p0:p1].T, kLI[p0:p1].T, pqkR.T, pqkI.T,
-1, j3cR[k], j3cI[k], 1)
log.debug3(' p0:p1 = %d:%d memory = %s',
p0, p1, lib.current_memory()[0])
for k, ji in enumerate(adapted_ji_idx):
if is_zero(kpt) and gamma_point(adapted_kptjs[k]):
save('j3c/%d'%ji, j3cR[k], col0, col1)
else:
save('j3c/%d'%ji, j3cR[k]+j3cI[k]*1j, col0, col1)
for k, kpt in enumerate(uniq_kpts):
make_kpt(k)
feri.close()
