def grada(direcs,v='bo1_H-H',dft='siesta',batch=100):
''' variables like: bo1_C-H, boc5_C rosi_C-H boc1
'''
print('- grading ... ... ')
for m in direcs:
mol = m
rn = IRNN(libfile='ffield',direcs=direcs,
dft=dft,
batch_size=batch,
pkl=True)
rn.initialize()
rn.session(learning_rate=1.0e-4,method='AdamOptimizer')
fg = open('gradient.txt','w')
l = 'eang'
grad = rn.get_gradient(rn.__dict__[l][mol],rn.p[v])
print('- the gradient of %s_%s/%s is: ' %(l,mol,v),grad,file=fg)
fg.close()
# anglit = ['D_ang','thet','theta0','expang','f_7','f_8','EANG','EPEN','ETC']
anglit = ['sbo','f_8','EANG','EPEN','ETC']
angs = rn.angs
# angs = ['H-O-H']
for ang in angs:
if rn.nang[ang]>0:
for l in anglit:
grad = rn.get_gradient(rn.__dict__[l][ang],rn.p[v])
fg = open('gradient.txt','a')
print('- the gradient of %s/%s is: ' %(l+'_'+ang,v),
grad,file=fg)
fg.close()
rn.sess.close()
def torsion(direcs=None,
dft='siesta'):
for m in direcs:
mol = m
rn = IRNN(libfile='ffield',direcs=direcs,
dft=dft,
rc_scale='none',
optword='all',
batch_size=200,
sort=False,
pkl=True,
interactive=True)
rn.initialize()
rn.session(learning_rate=1.0-4,method='AdamOptimizer')
# rn.plot()
BOtij = rn.get_value(rn.BOtij)
BOtjk = rn.get_value(rn.BOtjk)
BOtkl = rn.get_value(rn.BOtkl)
f10 = rn.get_value(rn.f_10)
f11 = rn.get_value(rn.f_11)
Etor= rn.get_value(rn.Etor)
torp = rn.torp
tors = rn.tors
# print('- shape of expv2: ',expv2['C'].shape)
print('- num of torp %d.' %len(torp))
print('- num of tors %d.' %len(tors))
spec = rn.spec
torlab= rn.lk.torlab
ntor = rn.ntor
tors = rn.tors
rn.sess.close()
for tn in tors:
if ntor[tn]>0:
for t in range(ntor[tn]):
if torlab[tn][t][0] == mol:
print('- ',t,tn,
'BOij:',BOtij[tn][t][0],
'BOjk:',BOtjk[tn][t][0],
'BOkl:',BOtkl[tn][t][0],
'f10:',f10[tn][t][0],
'f11:',f11[tn][t][0])
print('- shape of Etor:',Etor[mol].shape)
print('- num of torp %d.' %len(torp))
print('- num of tors %d.' %len(tors))
def allgrad(direcs=None,batch=100):
rn = IRNN(libfile='ffield',direcs=direcs,
dft='cpmd',
batch_size=batch,
pkl=True)
rn.initialize()
rn.session(learning_rate=1.0e-4,method='AdamOptimizer')
grads,v,vn = rn.get_all_gradient()
fg = open('grad.txt','w')
for g,t,tt in zip(grads,vn,v):
print('- gradients of %20s is: %12.5f' %(t,g),' value is:',tt,file=fg)
fg.close()
rn.sess.close()
def pldf(direcs={'ethane': '/home/feng/siesta/train2/ethane'},
val='bo2_C-C',
batch_size=1000):
for m in direcs:
mol = m
rn = IRNN(libfile='ffield.json',
direcs=direcs,
dft='siesta',
optword='all',
batch_size=batch_size,
rc_scale='none',
clip_op=False,
interactive=True)
molecules = rn.initialize()
rn.session(learning_rate=1.0e-10, method='AdamOptimizer')
df4 = rn.get_value(rn.df4)
bonds = rn.bonds
bd = 'C-H'
plt.figure() # temperature
plt.ylabel(r'$\Delta$ distribution')
plt.xlabel(r"The value of $\Delta$ minus bo")
nb_ = 100
hist, bin_ = np.histogram(df4[bd],
range=(-4.0, 1.0),
bins=nb_,
density=True)
plt.plot(bin_[:-1], hist, alpha=0.5, color='blue', label=r"$\Delta^{'}$")
plt.legend(loc='best')
plt.savefig('delta_minus_bo.eps')
plt.close()
def gradt(v='bo2_C-N'):
''' variables like: bo1_C-H, boc5_C rosi_C-H boc1
'''
print('- grading ... ... ')
direcs = {
'nm': '/home/feng/cpmd_data/packmol/nm',
'nme': '/home/feng/cpmd_data/packmol/nme'
}
rn = IRNN(libfile='ffield',
direcs=direcs,
dft='cpmd',
batch_size=200,
pkl=True)
rn.initialize()
rn.session(learning_rate=1.0e-4, method='AdamOptimizer')
fg = open('gradient.txt', 'w')
grad = rn.get_gradient(rn.Loss, rn.p[v])
print('- the gradient of Loss/%s is: ' % v, grad, file=fg)
molit = ['etor', 'efcon']
for mol in rn.direcs:
for l in molit:
grad = rn.get_gradient(rn.__dict__[l][mol], rn.p[v])
print('- the gradient of %s/%s is: ' % (l + '_' + mol, v),
grad,
file=fg)
fg.close()
torlit = ['f_10', 'f_11', 'expv2', 'ETOR', 'Efcon']
i = 0
for tor in rn.tors:
if rn.ntor[tor] > 0:
# if i<=100:
if tor == 'C-O-O-N' or tor == 'C-O-O-H':
for l in torlit:
fg = open('gradient.txt', 'a')
grad = rn.get_gradient(rn.__dict__[l][tor], rn.p[v])
print('- the gradient of %s/%s is: ' % (l + '_' + tor, v),
grad,
file=fg)
fg.close()
i += 1
rn.sess.close()
def delta(direcs={'ethane': '/home/gfeng/siesta/train/ethane'},
val='bo2_C-C',
batch_size=1):
for m in direcs:
mol = m
rn = IRNN(libfile='ffield.json',
direcs=direcs,
dft='siesta',
optword='all',
batch_size=batch_size,
rc_scale='none',
clip_op=False,
interactive=True)
molecules = rn.initialize()
rn.session(learning_rate=1.0e-10, method='AdamOptimizer')
delta = []
deltap = []
xs = np.linspace(0.9, 20.0, 50)
atom = 18
for x in xs:
rn.sess.run(tf.assign(rn.v[val], x))
D = rn.get_value(rn.D)
atlab = rn.lk.atlab
natom = molecules[mol].natom
d = np.zeros([natom, batch_size])
dp = np.zeros([natom, batch_size])
cell = rn.cell[mol]
Dp_ = {}
for sp in rn.spec:
if rn.nsp[sp] > 0:
Dp_[sp] = tf.gather_nd(rn.Deltap, rn.atomlist[sp])
Dp = rn.get_value(Dp_)
for sp in rn.spec:
if rn.nsp[sp] > 0:
for l, lab in enumerate(atlab[sp]):
if lab[0] == mol:
i = int(lab[1])
d[i] = D[sp][l]
dp[i] = Dp[sp][l]
delta.append(d[atom][0])
deltap.append(dp[atom][0])
# d[natom,nframe] dp[natom,nframe]
plt.figure() # temperature
plt.ylabel('Delta')
plt.xlabel(val)
plt.plot(xs, delta, color='red', alpha=0.2, label=r'Delta')
plt.plot(xs, deltap, color='blue', alpha=0.2, label='Delta\'')
plt.legend(loc='best')
plt.savefig('delta.eps')
plt.close()
def pldlc(direcs={'ch4c-1': '/home/feng/siesta/train2/ch4c1'},
val='bo2_C-C',
batch_size=200):
for m in direcs:
mol = m
rn = IRNN(libfile='ffield.json',
direcs=direcs,
dft='siesta',
optword='all',
batch_size=batch_size,
rc_scale='none',
clip_op=False,
interactive=True)
molecules = rn.initialize()
rn.session(learning_rate=1.0e-10, method='AdamOptimizer')
sp = 'O'
dlc = rn.get_value(rn.Delta_lpcorr)
dlp = rn.get_value(rn.Delta_lp)
# d[natom,nframe] dp[natom,nframe]
plt.figure() # temperature
plt.ylabel(r'$\Delta$ distribution')
plt.xlabel(r"The value of $\Delta$ and $\Delta'$")
nb_ = 500
sp = 'C'
hist, bin_ = np.histogram(dlp[sp],
range=(-3.0, 3.0),
bins=nb_,
density=True)
plt.plot(bin_[:-1],
hist,
alpha=0.5,
color='blue',
label=r"$\Delta^{lp}(%s)$ " % sp)
hist, bin_ = np.histogram(dlc[sp],
range=(-3.0, 3.0),
bins=nb_,
density=True)
plt.plot(bin_[:-1],
hist,
alpha=0.5,
color='yellowgreen',
label=r'$\Delta^{lc}$(%s)' % sp)
plt.legend(loc='best')
plt.savefig('deltalc_%s.eps' % sp)
plt.close()
def get_v(direcs={'ch4cdeb':'/home/feng/siesta/ch4c4'},
batch=50):
for m in direcs:
mol = m
rn = IRNN(libfile='ffield',direcs=direcs,
dft='siesta',
rc_scale='none',
optword='all',
batch_size=batch,
sort=False,
pkl=True,
interactive=True)
rn.initialize()
rn.session(learning_rate=1.0-4,method='AdamOptimizer')
p = rn.p_
Dp_ = {}
for sp in rn.spec:
if rn.nsp[sp]>0:
Dp_[sp] = tf.gather_nd(rn.Deltap,rn.atomlist[sp])
Dp = rn.get_value(Dp_)
Dpi = rn.get_value(rn.Dpi)
Dlp = rn.get_value(rn.Dlp)
powb = rn.get_value(rn.powb)
expb = rn.get_value(rn.expb)
bop_si = rn.get_value(rn.bop_si)
bosi = rn.get_value(rn.bosi)
bop_pi = rn.get_value(rn.bop_pi)
bopi = rn.get_value(rn.bopi)
bop_pp = rn.get_value(rn.bop_pp)
bopp = rn.get_value(rn.bopp)
Fsi = rn.get_value(rn.Fsi)
Fpi = rn.get_value(rn.Fpi)
Fpp = rn.get_value(rn.Fpp)
fsi_1 = rn.get_value(rn.fsi_1)
fpi_1 = rn.get_value(rn.fpi_1)
fpp_1 = rn.get_value(rn.fpp_1)
fsi_4 = rn.get_value(rn.fsi_4)
fpi_4 = rn.get_value(rn.fpi_4)
fpp_4 = rn.get_value(rn.fpp_4)
f_2 = rn.get_value(rn.f_2)
f_3 = rn.get_value(rn.f_3)
Delta_lpcorr = rn.get_value(rn.Delta_lpcorr)
Delta_lp = rn.get_value(rn.Delta_lp)
D = rn.get_value(rn.D)
slp = rn.get_value(rn.slp)
DPIL = rn.get_value(rn.DPIL)
DLP = rn.get_value(rn.DLP)
otrm1 = rn.get_value(rn.otrm1)
EOV = rn.get_value(rn.EOV)
EUN = rn.get_value(rn.EUN)
nbd = rn.nbd
nsp = rn.nsp
spec = rn.spec
bonds = rn.bonds
bd = 'H-H'
# bonds = [bd]
bdlab = rn.lk.bdlab
atlab = rn.lk.atlab
atlall = rn.lk.atlall
alist = rn.atomlist
fbo = open('bo.txt','w')
for bd in bonds:
if nbd[bd]>0:
for i,pb in enumerate(powb[bd]):
print('-bond %s:' %bd,
'-bosi- %10.8f' %bosi[bd][i][0],
'-bopi- %10.8f' %bopi[bd][i][0],
'-bopp- %10.8f' %bopp[bd][i][0],
'-bopsi- %10.8f' %bop_si[bd][i][0],
'-boppi- %10.8f' %bop_pi[bd][i][0],
'-boppp- %10.8f' %bop_pp[bd][i][0],
'-fsi_1- %11.8f' %fsi_1[bd][i][0],
'-fpi_1- %11.8f' %fpi_1[bd][i][0],
'-fpp_1- %11.8f' %fpp_1[bd][i][0],
'-fsi_4- %11.8f' %fsi_1[bd][i][0],
'-fpi_4- %11.8f' %fpi_1[bd][i][0],
'-fpp_4- %11.8f' %fpp_1[bd][i][0],
'-Fsi- %10.8f' %Fsi[bd][i][0],
'-Fpi- %10.8f' %Fpi[bd][i][0],
'-Fpp- %10.8f' %Fpp[bd][i][0],file=fbo)
fbo.close()
fa = open('a.txt','w')
# spec=['H']
for atom in spec:
if nsp[atom]>0:
for i,dl in enumerate(Delta_lpcorr[atom]):
print('-atom %s:' %atom,
'-Delta- %11.8f' %(D[atom][i][0]),
'-Deltap- %11.8f' %(Dp[atom][i][0]),
'-Delta_lpcorr- %11.8f' %Delta_lpcorr[atom][i][0],
'-Delta_lc+v- %11.8f' %(Delta_lpcorr[atom][i][0]+p['val_'+atom]),
'-Delta_lp- %11.8f' %Delta_lp[atom][i][0],
'-EOV- %11.8f' %EOV[atom][i][0],
'-EUN- %11.8f' %EUN[atom][i][0],file=fa)
fa.close()
fab = open('ab.txt','w')
# mol = 'chw2-2-0'
for bd in bonds:
if nbd[bd]>0:
[a1,a2] = bd.split('-')
for i,pb in enumerate(powb[bd]):
blb = bdlab[bd][i]
# if blb[0]==mol:
mol = blb[0]
na1 = atlab[a1].index([mol,blb[1]])
na2 = atlab[a2].index([mol,blb[2]])
na1_ = alist[a1][na1][0]
na2_ = alist[a2][na2][0]
# print(DLP.shape)
print('-bond %s:' %bd, # '-atm %d %d:' %(na1_,na2_),
'-fsi_1- %11.8f' %fsi_1[bd][i][0],
'-fpi_1- %11.8f' %fpi_1[bd][i][0],
'-fpp_1- %11.8f' %fpp_1[bd][i][0],
'-f_2- %11.8f' %f_2[bd][i][0],
'-f_3- %11.8f' %f_3[bd][i][0],
'-DPIL- %11.8f' %(DPIL[na1_][0]),
'-DPIL- %11.8f' %(DPIL[na2_][0]),
'-Dlp- %11.8f' %(Dlp[a1][na1][0]),
'-Dlp- %11.8f' %(Dlp[a2][na2][0]),
file=fab)
fab.close()
def compare_eb(dire):
system('rm *.pkl')
mol = 'dp'
# direcs={mol:dire}
direcs={'dp':'/home/gfeng/cpmd/train/nmr/dop'}
batch= {'others':10}
rn = IRNN(libfile='ffield',direcs=direcs,
dft='cpmd',
rc_scale='bo1',
optword='all',
batch_size=10,
sort=False,
pkl=True,
interactive=True)
rn.initialize()
rn.session(learning_rate=1.0-4,method='AdamOptimizer')
dboci = rn.get_value(rn.Di_boc)
dbocj = rn.get_value(rn.Dj_boc)
dp = rn.get_value(rn.Dp)
deltap= rn.get_value(rn.Deltap)
bop = rn.get_value(rn.BOP)
powb = rn.get_value(rn.powb)
expb = rn.get_value(rn.expb)
sieng= rn.get_value(rn.sieng)
EBD= rn.get_value(rn.EBD)
ebd= rn.get_value(rn.ebond)
ebda = rn.get_value(rn.ebda)
bdlink = rn.bdlink
print('- shape of BOP: ',bop.shape)
print('- number of bond: ',rn.lk.nbond)
print('- number of atoms: ',len(rn.lk.atom_lab))
print('- shape of deltap: ',deltap.shape)
print('- shape of ebda: ',ebda[mol].shape)
nbd = rn.nbd
bdlab = rn.lk.bdlab
bdlall = rn.lk.bdlall
bonds = rn.bonds
rn.sess.close()
del rn
## another session
tc = cnn(libfile='ffield',direcs=direcs,
dft='cpmd',
rc_scale='bo1',
optword='all',
batch_size=batch,
sort=False,
pkl=True,
interactive=True)
tc.session(learning_rate=1.0e-4,method='AdamOptimizer')
natom = tc.M[mol].molecule.natom
nbond = tc.M[mol].molecule.nbond
AN = tc.M[mol].molecule.atom_name
V = tc.get_value(tc.P[mol]['val'])
Dp,Db = tc.get_value([tc.M[mol].Deltap,tc.M[mol].Delta_boc])
Dp = Dp+V
Powb,Expb,Sieng,EBOND = tc.get_value([tc.M[mol].powb,tc.M[mol].expb,
tc.M[mol].sieng,tc.M[mol].EBOND])
ebond = tc.get_value(tc.M[mol].ebond)
E = np.zeros([natom,natom])
for bd in bonds:
if nbd[bd]>0:
# print('- shape of new style: ',bo[bd].shape)
print('\n- bd: %s \n' %bd)
for nb in range(nbd[bd]):
if bdlab[bd][nb][0] == mol:
iatom = int(bdlab[bd][nb][1])
jatom = int(bdlab[bd][nb][2])
E[iatom][jatom] = EBD[bd][nb][0]
E[jatom][iatom] = EBD[bd][nb][0]
if abs(EBOND[0][iatom][jatom]-EBD[bd][nb][0])>0.0001:
print('- ',bd,iatom,jatom,
'powb:',Powb[0][iatom][jatom],powb[bd][nb][0],
'expb:',Expb[0][iatom][jatom],expb[bd][nb][0],
'sieng:',Sieng[0][iatom][jatom],sieng[bd][nb][0],
'EBOND:',EBOND[0][iatom][jatom],EBD[bd][nb][0])
e = np.sum(E)
Et= EBOND[0]
ee= np.sum(Et)
print('- ebond:',ebond[0],ebd[mol][0],'e:',0.5*e,0.5*ee)
print('- number of bond:',len(bdlink[mol]),nbond)
tc.sess.close()
def compare_bo(dire):
# mol = 'mol'
# direcs={mol:dire}
mol = 'mol'
direcs={mol:dire}
# direcs={'nm13':'/home/gfeng/cpmd/train/nmr/nm13',
# 'nm4':'/home/gfeng/cpmd/train/nmr/nm2004',
# 'nm11':'/home/gfeng/cpmd/train/nmr/nm11'}
batch= {'others':10}
rn = IRNN(libfile='ffield',direcs=direcs,
dft='cpmd',
rc_scale='bo1',
optword='all',
batch_size=10,
sort=False,
pkl=True,
interactive=True)
rn.initialize()
rn.session(learning_rate=1.0-4,method='AdamOptimizer')
ebondr = rn.get_value(rn.ebond[mol])
bo = rn.get_value(rn.bo)
bop = rn.get_value(rn.bop)
rbd = rn.get_value(rn.rbd)
bodiv1 = rn.get_value(rn.bodiv1)
bop_pi = rn.get_value(rn.bop_pi)
dboci= rn.get_value(rn.Di_boc)
dbocj= rn.get_value(rn.Dj_boc)
Dp = rn.get_value(rn.Dp)
BOP = rn.get_value(rn.BOP)
nbd = rn.nbd
bdlab = rn.lk.bdlab
bonds = rn.bonds
rn.sess.close()
del rn
## another session
tc = cnn(libfile='ffield',direcs=direcs,
dft='cpmd',
rc_scale='bo1',
optword='all',
batch_size=batch,
sort=False,
pkl=True,
interactive=True)
tc.session(learning_rate=1.0e-4,method='AdamOptimizer')
ebondt = tc.get_value(tc.M[mol].ebond)
r,BOp,BO,Bodiv1,Bodiv2,Bodiv3,BOp_pi = \
tc.get_value([tc.M[mol].r,tc.M[mol].BOp,tc.M[mol].BO,
tc.M[mol].bodiv1,tc.M[mol].bodiv2,tc.M[mol].bodiv3,
tc.M[mol].BOp_pi])
Db = tc.get_value(tc.M[mol].Delta_boc)
Vb = tc.get_value(tc.M[mol].P['valboc'])
for bd in bonds:
if nbd[bd]>0:
# print('- shape of new style: ',bo[bd].shape)
print('\n- bd: %s \n' %bd)
for nb in range(nbd[bd]):
if bdlab[bd][nb][0] == mol:
iatom = int(bdlab[bd][nb][1])
jatom = int(bdlab[bd][nb][2])
if abs(BO[0][iatom][jatom]-bo[bd][nb][0])>0.0001:
print('- ',bd,iatom,jatom,
'BOp:',BOp[0][iatom][jatom],bop[bd][nb][0],
'Dboci:',Db[0][iatom],dboci[bd][nb][0],
'Dbocj:',Db[0][jatom],dbocj[bd][nb][0],
'BO:',BO[0][iatom][jatom],bo[bd][nb][0],)
tc.sess.close()
print('- shape of BOp: ',BOp.shape)
print('- shape of Dp: ',Dp.shape)
def gradb(direcs,v='bo5',bd='C-O',dft='siesta',batch=100):
''' variables like: bo1_C-H, boc5_C rosi_C-H boc1
'''
v = v+'_'+bd
print('- grading ... ...')
for m in direcs:
mol = m
rn = IRNN(libfile='ffield',direcs=direcs,
dft=dft,
batch_size=batch,
pkl=True,
interactive=True)
rn.initialize()
rn.session(learning_rate=3.0e-4,method='AdamOptimizer')
fg = open('gradient.txt','w')
bdlit = ['bop_si','bop_pi','bop_pp',
'bosi','F',
'bopi',
'bopp',
'bo','bso',
'powb','expb','EBD']
# bdlit = ['bo','EBD']
gl = rn.get_gradient(rn.Loss,rn.p[v])
print('- the gradient of Loss/%s is:' %v,gl,file=fg)
bonds = rn.bonds
# bonds = [bd]
for b in bonds:
if rn.nbd[b]>0:
for l in bdlit:
grad = rn.get_gradient(rn.__dict__[l][b],rn.p[v])
print('- the gradient of %s/%s is: ' %(l+'_'+b,v),
grad,file=fg)
fg.close()
ml = ['ebond','elone','eover','eunder','eang',
'epen','tconj','etor','efcon','evdw','ehb']
# for mol in direcs:
for l in ml:
grad = rn.get_gradient(rn.__dict__[l][mol],rn.p[v])
fg = open('gradient.txt','a')
print('- the gradient of %s_%s/%s is: ' %(l,mol,v),grad,file=fg)
fg.close()
alit = ['Delta_lp','Delta_lpcorr','Dpi','Delta_e','nlp','slp','EL',
'EOV','so','otrm1','otrm2',
'EUN']
atoms = rn.spec
for l in alit:
# atoms = bd.split('-')
for a in atoms:
grad = rn.get_gradient(rn.__dict__[l][a],rn.p[v])
fg = open('gradient.txt','a')
print('- the gradient of %s_%s/%s is: ' %(l,a,v),grad,file=fg)
fg.close()
rn.sess.close()
def compare_f(dire):
mol = 'mol'
direcs={mol:dire}
batch= {'others':10}
rn = IRNN(libfile='ffield',direcs=direcs,
dft='cpmd',
rc_scale='bo1',
optword='all',
batch_size=10,
sort=False,
pkl=True,
interactive=True)
rn.initialize()
rn.session(learning_rate=1.0-4,method='AdamOptimizer')
rn.plot()
ebondr = rn.get_value(rn.ebond[mol])
bo = rn.get_value(rn.bo)
bop = rn.get_value(rn.bop)
rbd = rn.get_value(rn.rbd)
bodiv1 = rn.get_value(rn.bodiv1)
bop_pi = rn.get_value(rn.bop_pi)
f = rn.get_value(rn.F)
f11 = rn.get_value(rn.F_11)
f12 = rn.get_value(rn.F_12)
f45 = rn.get_value(rn.F_45)
f4 = rn.get_value(rn.f_4)
f5 = rn.get_value(rn.f_5)
dboci= rn.get_value(rn.Di_boc)
dbocj= rn.get_value(rn.Dj_boc)
Dp = rn.get_value(rn.Dp)
BOP = rn.get_value(rn.BOP)
print('- shape of BOP: ',BOP.shape)
print('- number of bond: ',rn.lk.nbond)
print('- number of atoms: ',len(rn.lk.atom_lab))
nbd = rn.nbd
bdlab = rn.lk.bdlab
bonds = rn.bonds
rn.sess.close()
del rn
## another session
tc = cnn(libfile='ffield',direcs=direcs,
dft='cpmd',
rc_scale='bo1',
optword='all',
batch_size=batch,
sort=False,
pkl=True,
interactive=True)
tc.session(learning_rate=1.0e-4,method='AdamOptimizer')
ebondt = tc.get_value(tc.M[mol].ebond)
r,BOp,BO,Bodiv1,Bodiv2,Bodiv3,BOp_pi = \
tc.get_value([tc.M[mol].r,tc.M[mol].BOp,tc.M[mol].BO,
tc.M[mol].bodiv1,tc.M[mol].bodiv2,tc.M[mol].bodiv3,
tc.M[mol].BOp_pi])
fbosi = tc.get_value(tc.M[mol].fbosi)
F,F_11,F_12,F_45,F_4,F_5 = \
tc.get_value([tc.M[mol].F,tc.M[mol].F_11,tc.M[mol].F_12,tc.M[mol].F_45,
tc.M[mol].f_4,tc.M[mol].f_5])
Db = tc.get_value(tc.M[mol].Delta_boc)
Vb = tc.get_value(tc.M[mol].P['valboc'])
print('- shape of F_11: ',F_11.shape)
print('- shape of Delta_boc: ',Db.shape)
for bd in bonds:
if nbd[bd]>0:
# print('- shape of new style: ',bo[bd].shape)
print('\n- bd: %s \n' %bd)
for nb in range(nbd[bd]):
if bdlab[bd][nb][0] == mol:
iatom = int(bdlab[bd][nb][1])
jatom = int(bdlab[bd][nb][2])
if abs(BO[0][iatom][jatom]-bo[bd][nb][0])>0.0001:
print('- ',bd,iatom,jatom,
'BO:',BO[0][iatom][jatom],bo[bd][nb][0],
'F:',F[0][iatom][jatom],f[bd][nb][0],
'F_11:',F_11[0][iatom][jatom],f11[bd][nb][0],
'F_12:',F_12[0][iatom][jatom],f12[bd][nb][0],
'F_45:',F_45[0][iatom][jatom],f45[bd][nb][0],
'F_4:',F_4[0][iatom][jatom],f4[bd][nb][0],
'F_5:',F_5[0][iatom][jatom],f5[bd][nb][0],
'Dboci:',Db[0][iatom],dboci[bd][nb][0],
'Dbocj:',Db[0][jatom],dbocj[bd][nb][0])
tc.sess.close()
def debug_plot():
mol = 'ch4'
direcs={mol:'/home/feng/siesta/train/ch4'}
batch= {'others':300}
rn = IRNN(libfile='ffield',direcs=direcs,
dft='cpmd',
rc_scale='bo1',
optword='all',
batch_size=300,
sort=False,
pkl=True,
interactive=True)
rn.initialize()
rn.session(learning_rate=1.0-4,method='AdamOptimizer')
# rn.plot()
ebondr = rn.get_value(rn.ebond[mol])
eloner = rn.get_value(rn.elone[mol])
eoverr = rn.get_value(rn.eover[mol])
eunderr= rn.get_value(rn.eunder[mol])
eangr = rn.get_value(rn.eang[mol])
epenr = rn.get_value(rn.epen[mol])
tconjr = rn.get_value(rn.tconj[mol])
etorr = rn.get_value(rn.etor[mol])
efcon = rn.get_value(rn.efcon[mol])
evdw = rn.get_value(rn.evdw[mol])
rn.sess.close()
del rn
tc = cnn(libfile='ffield',direcs=direcs,
dft='cpmd',
rc_scale='bo1',
optword='all',
batch_size=batch,
sort=False,
pkl=True,
interactive=True)
tc.session(learning_rate=1.0e-4,method='AdamOptimizer')
ebondt = tc.get_value(tc.M[mol].ebond)
elonet = tc.get_value(tc.M[mol].elone)
eovert = tc.get_value(tc.M[mol].eover)
eundert = tc.get_value(tc.M[mol].eunder)
eangt = tc.get_value(tc.M[mol].eang)
epent = tc.get_value(tc.M[mol].epen)
tconjt= tc.get_value(tc.M[mol].tconj)
etort= tc.get_value(tc.M[mol].etor)
etort= tc.get_value(tc.M[mol].etor)
fconj= tc.get_value(tc.M[mol].fconj)
evdwt= tc.get_value(tc.M[mol].evdw)
tc.sess.close()
plot(ebondr,ebondt,'ebond','mol')
plot(eloner,elonet,'elone','mol')
plot(eoverr,eovert,'eover','mol')
plot(eunderr,eundert,'eunder','mol')
plot(eangr,eangt,'eang','mol')
plot(epenr,epent,'epen','mol')
plot(tconjr,tconjt,'etcon','mol')
plot(etorr,etort,'etor','mol')
plot(efcon,fconj,'efcon','mol')
plot(evdw,evdwt,'evdw','mol')
def get_v(direcs={'ch4cdeb': '/home/feng/siesta/ch4c4'}, batch=50):
for m in direcs:
mol = m
rn = IRNN(libfile='ffield',
direcs=direcs,
dft='siesta',
rc_scale='none',
optword='all',
batch_size=batch,
sort=False,
pkl=True,
interactive=True)
rn.initialize()
rn.session(learning_rate=1.0 - 4, method='AdamOptimizer')
p = rn.p_
Dp_ = {}
for sp in rn.spec:
if rn.nsp[sp] > 0:
Dp_[sp] = tf.gather_nd(rn.Deltap, rn.atomlist[sp])
Dp = rn.get_value(Dp_)
Dpi = rn.get_value(rn.Dpi)
Dlp = rn.get_value(rn.Dlp)
powb = rn.get_value(rn.powb)
expb = rn.get_value(rn.expb)
bop_si = rn.get_value(rn.bop_si)
bosi = rn.get_value(rn.bosi)
bop_pi = rn.get_value(rn.bop_pi)
bopi = rn.get_value(rn.bopi)
bop_pp = rn.get_value(rn.bop_pp)
bopp = rn.get_value(rn.bopp)
Delta_lpcorr = rn.get_value(rn.Delta_lpcorr)
Delta_lp = rn.get_value(rn.Delta_lp)
D = rn.get_value(rn.D)
slp = rn.get_value(rn.slp)
DPIL = rn.get_value(rn.DPIL)
DLP = rn.get_value(rn.DLP)
otrm1 = rn.get_value(rn.otrm1)
EOV = rn.get_value(rn.EOV)
EUN = rn.get_value(rn.EUN)
nbd = rn.nbd
nsp = rn.nsp
spec = rn.spec
bonds = rn.bonds
bd = 'H-H'
# bonds = [bd]
bdlab = rn.lk.bdlab
atlab = rn.lk.atlab
atlall = rn.lk.atlall
alist = rn.atomlist
fbo = open('bo.txt', 'w')
for bd in bonds:
if nbd[bd] > 0:
for i, pb in enumerate(powb[bd]):
print('-bond %s:' % bd,
'-bosi- %10.8f' % bosi[bd][i][0],
'-bopi- %10.8f' % bopi[bd][i][0],
'-bopp- %10.8f' % bopp[bd][i][0],
'-bopsi- %10.8f' % bop_si[bd][i][0],
'-boppi- %10.8f' % bop_pi[bd][i][0],
'-boppp- %10.8f' % bop_pp[bd][i][0],
file=fbo)
fbo.close()
fa = open('a.txt', 'w')
# spec=['H']
for atom in spec:
if nsp[atom] > 0:
for i, dl in enumerate(Delta_lpcorr[atom]):
print('-atom %s:' % atom,
'-Delta- %11.8f' % (D[atom][i][0]),
'-Deltap- %11.8f' % (Dp[atom][i][0]),
'-Delta_lpcorr- %11.8f' % Delta_lpcorr[atom][i][0],
'-Delta_lc+v- %11.8f' %
(Delta_lpcorr[atom][i][0] + p['val_' + atom]),
'-Delta_lp- %11.8f' % Delta_lp[atom][i][0],
'-EOV- %11.8f' % EOV[atom][i][0],
'-EUN- %11.8f' % EUN[atom][i][0],
file=fa)
fa.close()
def pldd(direcs={'ethane': '/home/feng/siesta/ethane'},
val='bo2_C-C',
batch_size=1000):
for m in direcs:
mol = m
rn = IRNN(libfile='ffield.json',
direcs=direcs,
dft='siesta',
optword='all',
batch_size=batch_size,
rc_scale='none',
clip_op=False,
interactive=True)
molecules = rn.initialize()
rn.session(learning_rate=1.0e-10, method='AdamOptimizer')
delta = []
deltap = []
D = rn.get_value(rn.D)
atlab = rn.lk.atlab
natom = molecules[mol].natom
d = np.zeros([natom, batch_size])
dp = np.zeros([natom, batch_size])
cell = rn.cell[mol]
Dp_ = {}
for sp in rn.spec:
if rn.nsp[sp] > 0:
Dp_[sp] = tf.gather_nd(rn.Deltap, rn.atomlist[sp])
Dp = rn.get_value(Dp_)
# d[natom,nframe] dp[natom,nframe]
plt.figure() # temperature
plt.ylabel(r'$\Delta$ distribution')
plt.xlabel(r"The value of $\Delta$ and $\Delta'$")
nb_ = 500
sp = 'O'
hist, bin_ = np.histogram(Dp[sp],
range=(np.min(Dp[sp]), np.max(Dp[sp])),
bins=nb_,
density=True)
plt.plot(bin_[:-1],
hist,
alpha=0.5,
color='blue',
linestyle=':',
label=r"$\Delta^{'}(%s)$ " % sp)
histp, bin_ = np.histogram(D[sp],
range=(np.min(D[sp]), np.max(D[sp])),
bins=nb_,
density=True)
plt.plot(bin_[:-1],
histp,
alpha=0.5,
color='yellowgreen',
linestyle='-.',
label=r'$\Delta$(%s)' % sp)
plt.legend(loc='best')
plt.savefig('delta_%s.eps' % sp)
plt.close()
def plot_delta(direcs=None, batch_size=1, dft='siesta'):
for m in direcs:
mol = m
rn = IRNN(libfile='ffield.json',
direcs=direcs,
dft=dft,
opt=[],
optword='all',
batch_size=batch_size,
rc_scale='none',
interactive=True)
molecules = rn.initialize()
rn.session(learning_rate=1.0e-10, method='AdamOptimizer')
D = rn.get_value(rn.D)
Dlp = rn.get_value(rn.Dlp)
Dp_ = {}
for sp in rn.spec:
if rn.nsp[sp] > 0:
Dp_[sp] = tf.gather_nd(rn.Deltap, rn.atomlist[sp])
Dp = rn.get_value(Dp_)
atlab = rn.lk.atlab
traj = Trajectory('delta.traj', 'w')
trajp = Trajectory('deltap.traj', 'w')
trajlp = Trajectory('deltalp.traj', 'w')
natom = molecules[mol].natom
d = np.zeros([natom, batch_size])
dp = np.zeros([natom, batch_size])
dlp = np.zeros([natom, batch_size])
cell = rn.cell[mol]
for sp in rn.spec:
if rn.nsp[sp] > 0:
for l, lab in enumerate(atlab[sp]):
if lab[0] == mol:
i = int(lab[1])
d[i] = D[sp][l]
dp[i] = Dp[sp][l]
dlp[i] = Dlp[sp][l]
for nf in range(batch_size):
A = Atoms(symbols=molecules[mol].atom_name,
positions=molecules[mol].x[nf],
charges=d[:, nf],
cell=cell,
pbc=(1, 1, 1))
traj.write(A)
Ap = Atoms(symbols=molecules[mol].atom_name,
positions=molecules[mol].x[nf],
charges=dp[:, nf],
cell=cell,
pbc=(1, 1, 1))
trajp.write(Ap)
Alp = Atoms(symbols=molecules[mol].atom_name,
positions=molecules[mol].x[nf],
charges=dlp[:, nf],
cell=cell,
pbc=(1, 1, 1))
trajlp.write(Alp)
traj.close()
trajp.close()
trajlp.close()