def make_hdiag_csf (h1e, eri, norb, nelec, transformer, hdiag_det=None):
smult = transformer.smult
if hdiag_det is None:
hdiag_det = make_hdiag_det (None, h1e, eri, norb, nelec)
eri = ao2mo.restore(1, eri, norb)
tlib = wlib = 0
neleca, nelecb = _unpack_nelec (nelec)
min_npair, npair_csd_offset, npair_dconf_size, npair_sconf_size, npair_sdet_size = get_csdaddrs_shape (norb, neleca, nelecb)
_, npair_csf_offset, _, _, npair_csf_size = get_csfvec_shape (norb, neleca, nelecb, smult)
npair_econf_size = npair_dconf_size * npair_sconf_size
max_npair = min (neleca, nelecb)
ncsf_all = count_all_csfs (norb, neleca, nelecb, smult)
ndeta_all = cistring.num_strings(norb, neleca)
ndetb_all = cistring.num_strings(norb, nelecb)
ndet_all = ndeta_all * ndetb_all
hdiag_csf = np.ascontiguousarray (np.zeros (ncsf_all, dtype=np.float64))
hdiag_csf_check = np.ones (ncsf_all, dtype=np.bool)
for npair in range (min_npair, max_npair+1):
ipair = npair - min_npair
nconf = npair_econf_size[ipair]
ndet = npair_sdet_size[ipair]
ncsf = npair_csf_size[ipair]
if ncsf == 0:
continue
nspin = neleca + nelecb - 2*npair
csd_offset = npair_csd_offset[ipair]
csf_offset = npair_csf_offset[ipair]
hdiag_conf = np.ascontiguousarray (np.zeros ((nconf, ndet, ndet), dtype=np.float64))
det_addr = transformer.csd_mask[csd_offset:][:nconf*ndet]
if ndet == 1:
# Closed-shell singlets
assert (ncsf == 1)
hdiag_csf[csf_offset:][:nconf] = hdiag_det[det_addr.flat]
hdiag_csf_check[csf_offset:][:nconf] = False
continue
det_addra, det_addrb = divmod (det_addr, ndetb_all)
det_stra = np.ascontiguousarray (cistring.addrs2str (norb, neleca, det_addra).reshape (nconf, ndet, order='C'))
det_strb = np.ascontiguousarray (cistring.addrs2str (norb, nelecb, det_addrb).reshape (nconf, ndet, order='C'))
det_addr = det_addr.reshape (nconf, ndet, order='C')
hdiag_conf = np.ascontiguousarray (np.zeros ((nconf, ndet, ndet), dtype=np.float64))
hdiag_conf_det = np.ascontiguousarray (hdiag_det[det_addr], dtype=np.float64)
t1 = time.process_time ()
w1 = time.time ()
libcsf.FCICSFhdiag (hdiag_conf.ctypes.data_as (ctypes.c_void_p),
hdiag_conf_det.ctypes.data_as (ctypes.c_void_p),
eri.ctypes.data_as (ctypes.c_void_p),
det_stra.ctypes.data_as (ctypes.c_void_p),
det_strb.ctypes.data_as (ctypes.c_void_p),
ctypes.c_uint (norb), ctypes.c_uint (nconf), ctypes.c_uint (ndet))
tlib += time.process_time () - t1
wlib += time.time () - w1
umat = get_spin_evecs (nspin, neleca, nelecb, smult)
hdiag_conf = np.tensordot (hdiag_conf, umat, axes=1)
hdiag_conf *= umat[np.newaxis,:,:]
hdiag_csf[csf_offset:][:nconf*ncsf] = hdiag_conf.sum (1).ravel (order='C')
hdiag_csf_check[csf_offset:][:nconf*ncsf] = False
assert (np.count_nonzero (hdiag_csf_check) == 0), np.count_nonzero (hdiag_csf_check)
#print ("Time in hdiag_csf library: {}, {}".format (tlib, wlib))
return hdiag_csf
def make_hdiag_csf_slower (h1e, eri, norb, nelec, transformer, hdiag_det=None):
''' This is tricky because I need the diagonal blocks for each configuration in order to get
the correct csf hdiag values, not just the diagonal elements for each determinant. '''
smult = transformer.smult
t0, w0 = time.process_time (), time.time ()
tstr = tlib = tloop = wstr = wlib = wloop = 0
if hdiag_det is None:
hdiag_det = make_hdiag_det (None, h1e, eri, norb, nelec)
eri = ao2mo.restore(1, eri, norb)
neleca, nelecb = _unpack_nelec (nelec)
min_npair, npair_csd_offset, npair_dconf_size, npair_sconf_size, npair_sdet_size = get_csdaddrs_shape (norb, neleca, nelecb)
_, npair_csf_offset, _, _, npair_csf_size = get_csfvec_shape (norb, neleca, nelecb, smult)
npair_econf_size = npair_dconf_size * npair_sconf_size
max_npair = min (neleca, nelecb)
ncsf_all = count_all_csfs (norb, neleca, nelecb, smult)
ndeta_all = cistring.num_strings(norb, neleca)
ndetb_all = cistring.num_strings(norb, nelecb)
ndet_all = ndeta_all * ndetb_all
hdiag_csf = np.ascontiguousarray (np.zeros (ncsf_all, dtype=np.float64))
hdiag_csf_check = np.ones (ncsf_all, dtype=np.bool)
for npair in range (min_npair, max_npair+1):
ipair = npair - min_npair
nconf = npair_econf_size[ipair]
ndet = npair_sdet_size[ipair]
ncsf = npair_csf_size[ipair]
if ncsf == 0:
continue
nspin = neleca + nelecb - 2*npair
csd_offset = npair_csd_offset[ipair]
csf_offset = npair_csf_offset[ipair]
hdiag_conf = np.ascontiguousarray (np.zeros ((nconf, ndet, ndet), dtype=np.float64))
det_addr = transformer.csd_mask[csd_offset:][:nconf*ndet]
if ndet == 1:
# Closed-shell singlets
assert (ncsf == 1)
hdiag_csf[csf_offset:][:nconf] = hdiag_det[det_addr.flat]
hdiag_csf_check[csf_offset:][:nconf] = False
continue
umat = get_spin_evecs (nspin, neleca, nelecb, smult)
det_addra, det_addrb = divmod (det_addr, ndetb_all)
t1, w1 = time.process_time (), time.time ()
det_stra = cistring.addrs2str (norb, neleca, det_addra).reshape (nconf, ndet, order='C')
det_strb = cistring.addrs2str (norb, nelecb, det_addrb).reshape (nconf, ndet, order='C')
tstr += time.process_time () - t1
wstr += time.time () - w1
det_addr = det_addr.reshape (nconf, ndet, order='C')
diag_idx = np.diag_indices (ndet)
triu_idx = np.triu_indices (ndet)
ipair_check = 0
# It looks like the library call below is, itself, usually responsible for about 50% of the
# clock and wall time that this function consumes.
t1, w1 = time.process_time (), time.time ()
for iconf in range (nconf):
addr = det_addr[iconf]
assert (len (addr) == ndet)
stra = det_stra[iconf]
strb = det_strb[iconf]
t2, w2 = time.process_time (), time.time ()
libfci.FCIpspace_h0tril(hdiag_conf[iconf].ctypes.data_as(ctypes.c_void_p),
h1e.ctypes.data_as(ctypes.c_void_p),
eri.ctypes.data_as(ctypes.c_void_p),
stra.ctypes.data_as(ctypes.c_void_p),
strb.ctypes.data_as(ctypes.c_void_p),
ctypes.c_int(norb), ctypes.c_int(ndet))
tlib += time.process_time () - t2
wlib += time.time () - w2
#hdiag_conf[iconf][diag_idx] = hdiag_det[addr]
#hdiag_conf[iconf] = lib.hermi_triu(hdiag_conf[iconf])
for iconf in range (nconf): hdiag_conf[iconf] = lib.hermi_triu (hdiag_conf[iconf])
for iconf in range (nconf): hdiag_conf[iconf][diag_idx] = hdiag_det[det_addr[iconf]]
tloop += time.process_time () - t1
wloop += time.time () - w1
hdiag_conf = np.tensordot (hdiag_conf, umat, axes=1)
hdiag_conf = (hdiag_conf * umat[np.newaxis,:,:]).sum (1)
hdiag_csf[csf_offset:][:nconf*ncsf] = hdiag_conf.ravel (order='C')
hdiag_csf_check[csf_offset:][:nconf*ncsf] = False
assert (np.count_nonzero (hdiag_csf_check) == 0), np.count_nonzero (hdiag_csf_check)
#print ("Total time in hdiag_csf: {}, {}".format (time.process_time () - t0, time.time () - w0))
#print (" Loop: {}, {}".format (tloop, wloop))
#print (" Library: {}, {}".format (tlib, wlib))
#print (" Cistring: {}, {}".format (tstr, wstr))
return hdiag_csf