def __init__(self,distr=None,xgrid=None):
"""
Initialize either with given probability distribution 'distr(x)',
or give the distribution on grid (given grid or 0,1,2,... by default)
"""
try:
distr(0.0)
self.f=distr
except:
self.x=xgrid
if self.x is None:
self.x=range(len(distr))
self.distr=distr
self.f=SplineFunction(self.x,self.distr,k=1)
# put the probability distribution f(x) on grid
self.N=1000
self.lim=self.f.limits()
self.xgrid=np.linspace(self.lim[0],self.lim[1],self.N)
self.fvals=[self.f(z) for z in self.xgrid]
# construct the distribution function F(x)
cum=[0.0]
for i in range(self.N-1):
cum.append(cum[-1]+self.fvals[i]+self.fvals[i+1])
cum=np.array(cum)/2*(self.lim[1]-self.lim[0])/self.N
self.norm=cum[-1]
cum=cum/self.norm
self.F=SplineFunction(self.xgrid,cum,k=1)
def __init__(self, distr=None, xgrid=None):
"""
Initialize either with given probability distribution 'distr(x)',
or give the distribution on grid (given grid or 0,1,2,... by default)
"""
try:
distr(0.0)
self.f = distr
except:
self.x = xgrid
if self.x is None:
self.x = range(len(distr))
self.distr = distr
self.f = SplineFunction(self.x, self.distr, k=1)
# put the probability distribution f(x) on grid
self.N = 1000
self.lim = self.f.limits()
self.xgrid = np.linspace(self.lim[0], self.lim[1], self.N)
self.fvals = [self.f(z) for z in self.xgrid]
# construct the distribution function F(x)
cum = [0.0]
for i in range(self.N - 1):
cum.append(cum[-1] + self.fvals[i] + self.fvals[i + 1])
cum = np.array(cum) / 2 * (self.lim[1] - self.lim[0]) / self.N
self.norm = cum[-1]
cum = cum / self.norm
self.F = SplineFunction(self.xgrid, cum, k=1)
def append_energy_slope(self,
weight,
p,
dEdp,
p0,
calc,
traj,
comment=None,
label=None,
color=None):
"""
Calculates the V'rep(r) at one point using trajectory over parameters p.
Trajectory is calculated using parameters p, giving E(p), where E is the total energy
without Vrep(r). The pair distance R=R(p). At p=p0, we set dE/dp|p=p0=dEdp, from which
we can set V'rep(R(p)) as
dEdp - dE/dp(p0)
V'rep(r) = -------------------
N * dR/dp
parameters:
===========
weight: fitting weight
p: parameter list
dEdp: slope of energy at p0
p0: the point where energy slope is set
calc: Hotbit calculator (remember charge and k-points)
traj: filename for ASE trajectory, or PickleTrajectory
object
comment: fitting comment for par-file (replaced by comment if None)
label: plotting label (replaced by comment if None)
color: plotting color
"""
raise NotImplementedError('This method was never tested properly.')
from box.interpolation import SplineFunction
R, E, N = [], [], []
for atoms in traj:
a, c = self._set_calc(atoms, calc)
e = a.get_potential_energy()
r, n = self._get_repulsion_distances(c)
if n > 0 and r < self.r_cut:
E.append(atoms.get_potential_energy())
R.append(r)
N.append(n)
R, E, N = np.array(R), np.array(E), np.array(N)
if np.any(N[0] != N):
raise NotImplementedError(
'The number of bonds changes during trajectory; check implementation.'
)
Ef = SplineFunction(p, E)
Rf = SplineFunction(p, R)
color = self._get_color(color)
comment += ';w=%.1f;N=%i' % (weight, N[0])
return self.append_point(weight, Rf(p0), (dEdp - Ef(p0, der=1)) /
(N[0] * Rf(p0, der=1)), comment, label, color)
class ArbitraryDistribution:
""" Class for generating random numbers form given distribution. """
def __init__(self,distr=None,xgrid=None):
"""
Initialize either with given probability distribution 'distr(x)',
or give the distribution on grid (given grid or 0,1,2,... by default)
"""
try:
distr(0.0)
self.f=distr
except:
self.x=xgrid
if self.x is None:
self.x=range(len(distr))
self.distr=distr
self.f=SplineFunction(self.x,self.distr,k=1)
# put the probability distribution f(x) on grid
self.N=1000
self.lim=self.f.limits()
self.xgrid=np.linspace(self.lim[0],self.lim[1],self.N)
self.fvals=[self.f(z) for z in self.xgrid]
# construct the distribution function F(x)
cum=[0.0]
for i in range(self.N-1):
cum.append(cum[-1]+self.fvals[i]+self.fvals[i+1])
cum=np.array(cum)/2*(self.lim[1]-self.lim[0])/self.N
self.norm=cum[-1]
cum=cum/self.norm
self.F=SplineFunction(self.xgrid,cum,k=1)
def __call__(self):
""" Generate random number from given distribution. """
rnd=np.random.uniform(0.0,1.0)
return self.F.solve(rnd)
def verify(self,nr=10000):
""" Plot and compare given distribution and simulated one. """
c=np.array([self() for rnd in range(nr)])
import pylab as pl
pl.hist(c,len(self.xgrid),normed=True)
distr=[self.f(x)/self.norm for x in self.xgrid]
pl.plot(self.xgrid,distr,c='r')
pl.show()
def run(self):
"""
Solve the self-consistent potential.
"""
self.timer.start('solve ground state')
print('\nStart iteration...', file=self.txt)
self.enl={}
self.d_enl={}
for n,l,nl in self.list_states():
self.enl[nl]=0.0
self.d_enl[nl]=0.0
N=self.grid.get_N()
# make confinement and nuclear potentials; intitial guess for veff
self.conf=array([self.confinement_potential(r) for r in self.rgrid])
self.nucl=array([self.V_nuclear(r) for r in self.rgrid])
self.get_veff_and_dens()
self.calculate_Hartree_potential()
#self.Hartree=np.zeros((N,))
for it in range(self.itmax):
self.veff=self.mix*self.calculate_veff()+(1-self.mix)*self.veff
if self.scalarrel:
veff = SplineFunction(self.rgrid, self.veff)
self.dveff = array([veff(r, der=1) for r in self.rgrid])
d_enl_max, itmax=self.solve_eigenstates(it)
dens0=self.dens.copy()
self.dens=self.calculate_density()
diff=self.grid.integrate(np.abs(self.dens-dens0),use_dV=True)
if diff<self.convergence['density'] and d_enl_max<self.convergence['energies'] and it > 5:
break
self.calculate_Hartree_potential()
if np.mod(it,10)==0:
print('iter %3i, dn=%.1e>%.1e, max %i sp-iter' %(it,diff,self.convergence['density'],itmax), file=self.txt)
if it==self.itmax-1:
if self.timing:
self.timer.summary()
raise RuntimeError('Density not converged in %i iterations' %(it+1))
self.txt.flush()
self.calculate_energies(echo=True)
print('converged in %i iterations' %it, file=self.txt)
print('%9.4f electrons, should be %9.4f' %(self.grid.integrate(self.dens,use_dV=True),self.nel), file=self.txt)
for n,l,nl in self.list_states():
self.Rnl_fct[nl]=Function('spline',self.rgrid,self.Rnlg[nl])
self.unl_fct[nl]=Function('spline',self.rgrid,self.unlg[nl])
self.timer.stop('solve ground state')
self.timer.summary()
self.txt.flush()
self.solved=True
if self.write != None:
f=open(self.write,'wb')
pickle.dump(self.rgrid, f)
pickle.dump(self.veff, f)
pickle.dump(self.dens, f)
f.close()
class ArbitraryDistribution:
""" Class for generating random numbers form given distribution. """
def __init__(self, distr=None, xgrid=None):
"""
Initialize either with given probability distribution 'distr(x)',
or give the distribution on grid (given grid or 0,1,2,... by default)
"""
try:
distr(0.0)
self.f = distr
except:
self.x = xgrid
if self.x is None:
self.x = range(len(distr))
self.distr = distr
self.f = SplineFunction(self.x, self.distr, k=1)
# put the probability distribution f(x) on grid
self.N = 1000
self.lim = self.f.limits()
self.xgrid = np.linspace(self.lim[0], self.lim[1], self.N)
self.fvals = [self.f(z) for z in self.xgrid]
# construct the distribution function F(x)
cum = [0.0]
for i in range(self.N - 1):
cum.append(cum[-1] + self.fvals[i] + self.fvals[i + 1])
cum = np.array(cum) / 2 * (self.lim[1] - self.lim[0]) / self.N
self.norm = cum[-1]
cum = cum / self.norm
self.F = SplineFunction(self.xgrid, cum, k=1)
def __call__(self):
""" Generate random number from given distribution. """
rnd = np.random.uniform(0.0, 1.0)
return self.F.solve(rnd)
def verify(self, nr=10000):
""" Plot and compare given distribution and simulated one. """
c = np.array([self() for rnd in range(nr)])
import pylab as pl
pl.hist(c, len(self.xgrid), normed=True)
distr = [self.f(x) / self.norm for x in self.xgrid]
pl.plot(self.xgrid, distr, c='r')
pl.show()
def derivative(self, f, order=1):
"""
Get order-th derivative of function f (given with N grid points).
"""
x = self.get_grid()
f_spline = SplineFunction(x, f)
dfdx = f_spline(x, der=order)
return dfdx
def _integrate_vrep(self, dv_rep, r_cut, N=100):
"""
Integrate V'_rep(r) from r_cut to zero to get the V_rep(r)
"""
from box.interpolation import SplineFunction
from scipy.integrate import quadrature
r_g = np.linspace(r_cut, 0, N)
dr = r_g[1] - r_g[0]
v_rep = np.zeros(N)
for i in range(1,len(r_g)):
v_rep[i] = v_rep[i-1]
val, err = quadrature(dv_rep, r_g[i-1], r_g[i], tol=1.0e-12, maxiter=50)
v_rep[i] += val
# SplineFunction wants the x-values in ascending order
return SplineFunction(r_g[::-1], v_rep[::-1])
def vrep_to_spline_coefficients(rep):
"""
Transform simple grid data to spline form used by DFTB+
Input:
rep: rep[:,0]=r and rep[:,1]=v(r)
Output:
spline: List with spline data
Each line with: start end c0 c1 c2 c3
where v is interpolated between [start,end] using
v(r) = c0 + c1*(r-start)+ c2*(r-start)**2 + c3*(r-start)**3
The last line has coefficients up to c5.
"""
print(rep.shape)
v = SplineFunction(rep[:, 0], rep[:, 1])
n = rep.shape[0]
m = 6
spline = []
for i in range(n - 1):
r0, r1 = rep[i, 0], rep[i + 1, 0]
v0, v1 = rep[i, 1], rep[i + 1, 1]
rlist = linspace(r0, r1, m)
vlist = v(rlist)
if i == n - 2:
nc = 1 + 5
else:
nc = 1 + 3
pguess = zeros(nc)
pguess[0] = 0.5 * (v0 + v1)
pguess[1] = (v1 - v0) / (r1 - r0)
def chi2(p, x, y):
err = y - vrep_poly(x, r0, p)
return sum(err**2)
p = fmin(chi2, pguess, args=(rlist, vlist), xtol=1E-12, disp=False)
spline.append([r0, r1] + list(p))
return spline
def append_energy_curve(self,weight,calc,traj,comment=None,label=None,color=None):
"""
Calculates the V'rep(r) from a given ase-trajectory.
The trajectory can be anything, as long as the ONLY missing energy
from DFTB calculation is N*V_rep(R). Hence
E_DFT(R) = E_wr(R) + N*V_rep(R)
E_DFT'(R) - E_wr'(R)
V_rep'(R) = ------------------ ,
N
where R is the nn. distance,N is the number of A-B pairs taken into account,
and E_wr(R) = E_bs(R) + E_coul(R) is the DFTB energy without repulsion.
At least 3 points in energy curve needed, preferably more.
parameters:
===========
weight: fitting weight
calc: Hotbit calculator (remember charge and k-points)
traj: filename for ASE trajectory, or PickleTrajectory
object
comment: fitting comment for par-file (replaced by comment if None)
label: plotting label (replaced by comment if None)
color: plotting color
"""
if comment==None: comment=label
if label==None: label=comment
#if not ( isinstance(traj, type(PickleTrajectory)) or isinstance(traj, list) ):
if not ( isinstance(traj, type(PickleTrajectory)) or isinstance(traj, list) ):
print("\nAppending energy curve data from %s..." %traj, file=self.txt)
traj = PickleTrajectory(traj)
else:
print('\nAppending energy curve data...', file=self.txt)
Edft, Ewr, N, R = [], [], [], []
if len(traj)<3:
raise AssertionError('At least 3 points in energy curve required.')
for atoms in traj:
a, c = self._set_calc(atoms,calc)
e = a.get_potential_energy()
r, n = self._get_repulsion_distances(c)
if n>0 and r<self.r_cut:
Edft.append( atoms.get_potential_energy() )
Ewr.append( e )
R.append(r)
N.append(n)
Edft = np.array(Edft)
Ewr = np.array(Ewr)
N = np.array(N)
R = np.array(R)
if np.any( N-N[0]!=0 ):
raise RuntimeError('The number of bonds changes within trajectory.')
# sort radii because of spline
ind = R.argsort()
R = R[ind]
Edft = Edft[ind]
Ewr = Ewr[ind]
from box.interpolation import SplineFunction
k = min(len(Edft)-2,3)
vrep = SplineFunction(R, (Edft-Ewr)/N, k=k, s=0)
color = self._get_color(color)
for i, r in enumerate(R):
if i==0:
com = comment + ';w=%.1f' %weight
else:
label='_nolegend_'
com = None
self.append_point(weight/np.sqrt(len(R)),r, vrep(r,der=1), com, label, color)
print("Appended %i points around R=%.4f...%.4f" %(len(N),R.min(),R.max()), file=self.txt)
def write_skf(el1, el2, data1, data2, tables, rep, spline, dr=0.1):
""" Write .skf files for el1-el2 interactions.
Input:
el1: element 1 symbol
el2: element 2 symbol
data1: dictionary for element 1 data
data2: dictionary for element 2 data
tables: dictionary for Slater-Koster table data
rep: repulsive potential on a grid
spline: spline in appropriate format already
dr: output grid spacing
"""
if el1 == el2:
filename = '%s-%s.skf' % (el1, el2)
o = open(filename, 'w')
ng = int(ceil(tables['grid'][-1] / dr))
o.write('%.15f, %i, ' % (dr, ng))
e = data1['energies']
u = data1['U']
occ = data1['occupations']
occ = [2, 2]
# Ed Ep Es SPE Ud Up Us fd fp fs (orbital energies and Hubbard U)
o.write("{} {} {} {} {} {} {} {} {} {} \n".format(
0.0, e[0], e[1], 0., u, u, u, 0.0, occ[0], occ[1]))
# mass c2 c3 c4 c5 c6 c7 c8 c9 rcut d1 d2 d3 d4 d5 d6 d7 d8 d9 d10
# where c* are vrep coefficients (unless spline) and d* are not used
o.write("{} {} {} {} {} {} {} {} {} {} \n".format(
data1['mass'] * 2, 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0.))
#
# interpolate SlaKo tables
#
tab = tables['%s%s' % (el1, el2)] #[i,j]
f = [SplineFunction(tables['grid'], tab[:, i]) for i in range(20)]
for i in range(ng):
for j in range(20):
o.write("{} ".format(f[j]((i + 1) * dr)))
o.write("\n")
#print>>o
#
# repulsion
#
o.write('Spline \n')
o.write("{} {} \n".format(len(spline), spline[-1][1])) #nInt, cutoff
o.write(" {} {} {} \n".format(0, -1E19, spline[0][2]))
for sp in spline:
#print(' '.join([str(x) for x in sp]))
o.write(' '.join([str(x) for x in sp]))
o.write("\n")
#print>>o
o.close()
else:
for e1, e2 in [(el1, el2), (el2, el1)]:
filename = '%s-%s.skf' % (e1, e2)
o = open(filename, 'w')
ng = int(ceil(tables['grid'][-1] / dr))
o.write('%.15f, %i, ' % (dr, ng))
o.write(0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0.)
#
# interpolate SlaKo tables
#
tab = tables['%s%s' % (e1, e2)] #[i,j]
f = [SplineFunction(tables['grid'], tab[:, i]) for i in range(20)]
for i in range(ng):
for j in range(20):
o.write(f[j]((i + 1) * dr), '\n')
#print>>o
#
# repulsion
#
o.write('Spline')
o.write(len(spline), spline[-1][1]) #nInt, cutoff
o.write(0, -1E19, spline[0][2])
for sp in spline:
o.write(' '.join([str(x) for x in sp], '\n'))
#print>>o
o.close()
