def __init__(self, brlu, ltris, ladjs, mpinfo, mlpats, \
mdisp, mdispgd, ksft, lamb0, find_mode):
self._brlu = brlu
self._ltris = ltris
self._nps = numpy.sqrt(len(ltris)) / 2
self._bandnum = len(mpinfo)
self._ladjs = ladjs
self._mpinfo = mpinfo
self._mlpats = mlpats
self._mdisp = mdisp
self._mdispgd = mdispgd
self._ksft = ksft
self._lamb0 = lamb0
self._patchnum = len(self._mpinfo[0])
#找到每个对应的n4
#现在每个带都有自己的idx4
if find_mode == 1:
data_list = []
idxit = numpy.nditer(U, flags=['multi_index'])
step = numpy.minimum(brlu.width, brlu.height) / 10 / self._nps
while not idxit.finished:
bd1, bd2, bd3, bd4, idx1, idx2, idx3 = idxit.multi_index
kv1, kv2, kv3 = mpinfo[bd1, idx1], mpinfo[bd2,
idx2], mpinfo[bd3,
idx3]
kv4 = ksft(ksft(kv1, kv2),
Point(-kv3.coord[0], -kv3.coord[1], 1))
data_list.append(
(kv4, mpinfo[bd4, :], mdisp[bd4], mdispgd[bd4], step))
#idx4 = find_patch(kv4, mpinfo[bd4, :], mdisp[bd4],\
# mdispgd[bd4], numpy.pi / 2 / self._nps)
#self._mk4tab[idxit.multi_index] = idx4
idxit.iternext()
with multiprocessing.Pool(get_procs_num()) as pool:
self._mk4tab = pool.starmap(find_patch, data_list)
self._mk4tab = numpy.reshape(self._mk4tab, U.shape)
return
data_list = []
idxit = numpy.nditer(U, flags=['multi_index'])
while not idxit.finished:
bd1, bd2, bd3, bd4, idx1, idx2, idx3 = idxit.multi_index
kv1, kv2, kv3 = mpinfo[bd1, idx1], mpinfo[bd2, idx2], mpinfo[bd3,
idx3]
kv4 = ksft(ksft(kv1, kv2), Point(-kv3.coord[0], -kv3.coord[1], 1))
data_list.append((kv4, mpinfo[bd4, :]))
idxit.iternext()
find_func = {2: find_patch_mode2, 3: find_patch_mode3}[find_mode]
with multiprocessing.Pool(get_procs_num()) as pool:
self._mk4tab = pool.starmap(find_func, data_list)
self._mk4tab = numpy.reshape(self._mk4tab, U.shape)
def slove_equ(args, brlu, ltris, ladjs, mpinfo, mlpats):
'''解方程'''
#初始化U
uval = numpy.zeros((1, 1, 1, 1, args.patches, args.patches, args.patches))
uval[0, 0, 0, 0, :, :, :] += band_u(6.0, mpinfo[0])
uval[0, 0, 0, 0, :, :, :] += band_v(6.0, mpinfo[0])
#初始化hubbard
hubbard.uinit(uval, args.patches, 1)
lamb0 = 2.0
print('lamb0 = ', lamb0)
hubbard.config_init(brlu,
ltris,
ladjs,
mpinfo,
mlpats, [p_disp], [None],
shift_kv,
lamb0,
find_mode=3)
#输出文件夹
if not os.path.isdir('heatmap8'):
os.mkdir('heatmap8')
rpath = 'heatmap8/kagome4'
if not os.path.isdir(rpath):
os.mkdir(rpath)
lval = 0.
lstep = 0.01
numpy.save('{0}/{1:.2f}U.npy'.format(rpath, lval), hubbard.U)
#读取配置
#lval = 6.70
#hubbard.U = numpy.load('{0}/{1:.2f}U.npy'.format(rpath, lval))
for _ in range(1000):
hubbard.precompute_contour(lval)
hubbard.precompute_qpp(lval)
hubbard.precompute_qfs(lval)
hubbard.precompute_nqfs(lval)
hubbard.precompute_qex(lval)
hubbard.precompute_nqex(lval)
#计算每个idx的导数
data_list = []
ndit = numpy.nditer(hubbard.U, flags=['multi_index'])
while not ndit.finished:
bd1, bd2, bd3, bd4, idx1, idx2, idx3 = ndit.multi_index
data_list.append((lval, bd1, bd2, bd3, bd4, idx1, idx2, idx3))
ndit.iternext()
#进程池
#KNOWN ISSUE: 在修改全局变量之间建立的Pool，里面不会包含全局变量
with multiprocessing.Pool(get_procs_num()) as pool:
result = pool.starmap(hubbard.dl_ec, data_list)
duval = numpy.reshape(result, hubbard.U.shape)
#把每个idx的值加上
#这两个过程不能放在一起，因为计算dl_ec的时候用到了hubbard.U
hubbard.U += duval * lstep
lval += lstep
#
del data_list, result, duval
#
#uval2 = numpy.load('{0}/{1:.2f}U.chk'.format(rpath, 10.76))
#if lval == 10.76:
# assert numpy.allclose(hubbard.U, uval2)
numpy.save('{0}/{1:.2f}U.npy'.format(rpath, lval), hubbard.U)
def slove_equ(args, brlu, ltris, ladjs, pinfo, lpats):
'''解方程'''
disp = {'square': dispersion, 'hole': hole_disp}[args.disp]
dispgd = {
'square': dispersion_gradient,
'hole': dispersion_gradient
}[args.disp]
#初始化U
hubbard.uinit(1.0, args.patches)
#初始化hubbard模型
hubbard.config_init(brlu, ltris, ladjs, pinfo, lpats, disp, dispgd,
shift_kv, 4.0)
#输出文件夹
if not os.path.isdir('heatmap5'):
os.mkdir('heatmap5')
lval = 0.
lstep = 0.01
draw_heatmap(hubbard.U[:, :, 0], save='heatmap5/{:.2f}.jpg'.format(lval))
for _ in range(320):
#duval = numpy.zeros_like(hubbard.U)
hubbard.precompute_contour(lval)
hubbard.precompute_qpp(lval)
hubbard.precompute_qfs(lval)
hubbard.precompute_nqfs(lval)
hubbard.precompute_qex(lval)
hubbard.precompute_nqex(lval)
#计算每个idx的导数
data_list = []
for idx1 in range(args.patches):
for idx2 in range(args.patches):
for idx3 in range(args.patches):
data_list.append((lval, idx1, idx2, idx3))
#duval[idx1, idx2, idx3] = hubbard.dl_ec(lval, idx1, idx2, idx3)
#进程池
#KNOWN ISSUE: 在修改全局变量之间建立的Pool，里面不会包含全局变量
with multiprocessing.Pool(get_procs_num()) as pool:
result = pool.starmap(hubbard.dl_ec, data_list)
duval = numpy.reshape(result,
(args.patches, args.patches, args.patches))
#把每个idx的值加上
#这两个过程不能放在一起，因为计算dl_ec的时候用到了hubbard.U
hubbard.U += duval * lstep
lval += lstep
draw_heatmap(hubbard.U[:, :, 0],
save='heatmap5/{:.2f}.jpg'.format(lval))
def precompute_qfs(lval):
'''计算q_fs实际上也不需要多余的变量
bd2, bd3, idx2, idx3
'''
global QUICKQFS
#fs的所有自由度
data_list = []
place_holder = numpy.ndarray(( #alpha, beta, bd2, bd3
CONFIG.bandnum,
CONFIG.bandnum,
CONFIG.bandnum,
CONFIG.bandnum,
#nidx(alpha), idx2(bd2), idx3(bd3)
CONFIG.patchnum,
CONFIG.patchnum,
CONFIG.patchnum))
nditer = numpy.nditer(place_holder, flags=['multi_index'])
#
area = CONFIG.brlu.width * CONFIG.brlu.height
ltris = CONFIG.ltris
tarea = area / len(ltris)
mpinfo = CONFIG.mpinfo
ksft = CONFIG.ksft
lamb = CONFIG.lamb0 * numpy.exp(-lval)
mdisp = CONFIG.mdisp
#
while not nditer.finished:
alpha, beta, bd2, bd3, nidx, idx2, idx3 = nditer.multi_index
kv2, kv3 = mpinfo[bd2, idx2], mpinfo[bd3, idx3]
q_fs = ksft(kv3, Point(-kv2.coord[0], -kv2.coord[1], 1))
data_list.append(
(PATTRIS[alpha, nidx], tarea, lamb,\
mdisp[alpha], mdisp[beta], q_fs, ksft, area)
)
nditer.iternext()
#
with multiprocessing.Pool(get_procs_num()) as pool:
result = pool.starmap(pi_ab_plus_tf, data_list)
QUICKQFS = (lval, numpy.reshape(result, place_holder.shape))
def precompute_qpp(lval):
'''Pi^{-}实际上不需要7个指标，只需要
bd1, idx1, bd2, idx2
另外的alpha，beta，nidx在所有情况下都是需要遍历的，所以可以提前计算出来，节约计算时间
'''
global QUICKQPP
data_list = []
place_holder = numpy.ndarray(( #alpha, beta, bd1, bd2
CONFIG.bandnum,
CONFIG.bandnum,
CONFIG.bandnum,
CONFIG.bandnum,
#nidx(alpha), idx1(bd1), idx2(bd2)
CONFIG.patchnum,
CONFIG.patchnum,
CONFIG.patchnum))
nditer = numpy.nditer(place_holder, flags=['multi_index'])
#
area = CONFIG.brlu.width * CONFIG.brlu.height
ltris = CONFIG.ltris
tarea = area / len(ltris)
mpinfo = CONFIG.mpinfo
ksft = CONFIG.ksft
lamb = CONFIG.lamb0 * numpy.exp(-lval)
mdisp = CONFIG.mdisp
#
while not nditer.finished:
alpha, beta, bd1, bd2, nidx, idx1, idx2 = nditer.multi_index
kv1, kv2 = mpinfo[bd1, idx1], mpinfo[bd2, idx2]
q_pp = ksft(kv1, kv2)
#
data_list.append(
(PATTRIS[alpha, nidx], tarea, lamb,\
mdisp[alpha], mdisp[beta], q_pp, ksft, area)
)
nditer.iternext()
with multiprocessing.Pool(get_procs_num()) as pool:
result = pool.starmap(pi_ab_minus_tf, data_list)
QUICKQPP = (lval, numpy.reshape(result, place_holder.shape))
def slove_equ(args, brlu, ltris, ladjs, mpinfo, mlpats):
'''解方程'''
#初始化U
spats = mpinfo[0, :]
ppats = mpinfo[1, :]
hubbard.uinit(get_initu(spats, ppats, 2.0), args.patches, 2)
#初始化hubbard模型
lamb0 = get_max_val()
print('lamb0 = ', lamb0)
hubbard.config_init(brlu, ltris, ladjs, mpinfo, mlpats,
[s_band_disp, p_band_disp], [s_band_gd, p_band_gd],
shift_kv, lamb0)
#
#输出文件夹
if not os.path.isdir('heatmap6'):
os.mkdir('heatmap6')
rpath = 'heatmap6/s{0:.2f}nu{1:.2f}'.\
format(args.stripe, args.nu)
if not os.path.isdir(rpath):
os.mkdir(rpath)
#
lval = 0.
lstep = 0.01
numpy.save('{0}/{1:.2f}U.npy'.format(rpath, lval), hubbard.U)
print(hubbard.U[0, 0, 0, 0, 0, 0, 0])
print(hubbard.U[1, 1, 1, 1, 0, 0, 0])
print(hubbard.U[1, 1, 1, 0, 1, 1, 1])
print(hubbard.U[0, 1, 1, 1, 1, 1, 1])
print(hubbard.U[1, 1, 0, 0, 2, 2, 2])
print(hubbard.U[0, 0, 1, 1, 2, 2, 2])
print(hubbard.U[1, 1, 1, 0, 3, 3, 3])
print(hubbard.U[0, 1, 1, 1, 3, 3, 3])
#加载上一次的结果
#lval = 10.75
#hubbard.U = numpy.load('{0}/{1:.2f}U.npy'.format(rpath, lval))
for _ in range(1000):
hubbard.precompute_contour(lval)
hubbard.precompute_qpp(lval)
hubbard.precompute_qfs(lval)
hubbard.precompute_nqfs(lval)
hubbard.precompute_qex(lval)
hubbard.precompute_nqex(lval)
#计算每个idx的导数
data_list = []
ndit = numpy.nditer(hubbard.U, flags=['multi_index'])
while not ndit.finished:
bd1, bd2, bd3, bd4, idx1, idx2, idx3 = ndit.multi_index
data_list.append((lval, bd1, bd2, bd3, bd4, idx1, idx2, idx3))
ndit.iternext()
#进程池
#KNOWN ISSUE: 在修改全局变量之间建立的Pool，里面不会包含全局变量
with multiprocessing.Pool(get_procs_num()) as pool:
result = pool.starmap(hubbard.dl_ec, data_list)
duval = numpy.reshape(result, hubbard.U.shape)
#把每个idx的值加上
#这两个过程不能放在一起，因为计算dl_ec的时候用到了hubbard.U
hubbard.U += duval * lstep
lval += lstep
#
del data_list, result, duval
#
#uval2 = numpy.load('{0}/{1:.2f}U.chk'.format(rpath, 10.76))
#if lval == 10.76:
# assert numpy.allclose(hubbard.U, uval2)
numpy.save('{0}/{1:.2f}U.npy'.format(rpath, lval), hubbard.U)
print(hubbard.U[0, 0, 0, 0, 0, 0, 0])
print(hubbard.U[1, 1, 1, 1, 0, 0, 0])
print(hubbard.U[1, 1, 1, 0, 1, 1, 1])
print(hubbard.U[0, 1, 1, 1, 1, 1, 1])
print(hubbard.U[1, 1, 0, 0, 2, 2, 2])
print(hubbard.U[0, 0, 1, 1, 2, 2, 2])
print(hubbard.U[1, 1, 1, 0, 3, 3, 3])
print(hubbard.U[0, 1, 1, 1, 3, 3, 3])