def getXC(gr,nel,**kwargs):
"Form the exchange-correlation matrix"
functional = kwargs.get('functional','SVWN')
do_grad_dens = need_gradients[functional]
do_spin_polarized = kwargs.get('do_spin_polarized')
gr.floor_density() # Insure that the values of the density don't underflow
gr.renormalize(nel) # Renormalize to the proper # electrons
dens = gr.dens()
weight = gr.weights()
gamma = gr.get_gamma()
npts = len(dens)
if gr.version == 1:
amdens = zeros((2,npts),'d')
amgamma = zeros((3,npts),'d')
amdens[0,:] = amdens[1,:] = 0.5*dens
if gamma is not None:
amgamma[0,:] = amgamma[1,:] = amgamma[2,:] = 0.25*gamma
elif gr.version == 2:
amdens = gr.density.T
amgamma = gr.gamma.T
fxc,dfxcdna,dfxcdnb,dfxcdgaa,dfxcdgab,dfxcdgbb = XC(amdens,amgamma,**kwargs)
Exc = dot(weight,fxc)
wva = weight*dfxcdna # Combine w*v in a vector for multiplication by bfs
# First do the part that doesn't depend upon gamma
nbf = gr.get_nbf()
Fxca = zeros((nbf,nbf),'d')
for i in xrange(nbf):
wva_i = wva*gr.bfgrid[:,i]
for j in xrange(nbf):
Fxca[i,j] = dot(wva_i,gr.bfgrid[:,j])
# Now do the gamma-dependent part.
# Fxc_a += dot(2 dfxcdgaa*graddensa + dfxcdgab*graddensb,grad(chia*chib))
# We can do the dot product as
# sum_grid sum_xyz A[grid,xyz]*B[grid,xyz]
# or a 2d trace product
# Here A contains the dfxcdgaa stuff
# B contains the grad(chia*chib)
# Possible errors: gr.grad() here should be the grad of the b part?
if do_grad_dens:
# A,B are dimensioned (npts,3)
A = transpose(0.5*transpose(gr.grad())*(weight*(2*dfxcdgaa+dfxcdgab)))
for a in xrange(nbf):
for b in xrange(a+1):
B = gr.grad_bf_prod(a,b)
Fxca[a,b] += sum(ravel(A*B))
Fxca[b,a] = Fxca[a,b]
if not do_spin_polarized: return Exc,Fxca
wvb = weight*dfxcdnb # Combine w*v in a vector for multiplication by bfs
# First do the part that doesn't depend upon gamma
Fxcb = zeros((nbf,nbf),'d')
for i in xrange(nbf):
wvb_i = wvb*gr.bfgrid[:,i]
for j in xrange(nbf):
Fxcb[i,j] = dot(wvb_i,gr.bfgrid[:,j])
# Now do the gamma-dependent part.
# Fxc_b += dot(2 dfxcdgbb*graddensb + dfxcdgab*graddensa,grad(chia*chib))
# We can do the dot product as
# sum_grid sum_xyz A[grid,xyz]*B[grid,xyz]
# or a 2d trace product
# Here A contains the dfxcdgaa stuff
# B contains the grad(chia*chib)
# Possible errors: gr.grad() here should be the grad of the b part?
if do_grad_dens:
# A,B are dimensioned (npts,3)
A = transpose(0.5*transpose(gr.grad())*(weight*(2*dfxcdgbb+dfxcdgab)))
for a in xrange(nbf):
for b in xrange(a+1):
B = gr.grad_bf_prod(a,b)
Fxcb[a,b] += sum(ravel(A*B))
Fxcb[b,a] = Fxcb[a,b]
return Exc,Fxca,Fxcb
def getXC(gr, nel, **kwargs):
"Form the exchange-correlation matrix"
functional = kwargs.get('functional', 'SVWN')
do_grad_dens = need_gradients[functional]
do_spin_polarized = kwargs.get('do_spin_polarized')
gr.floor_density() # Insure that the values of the density don't underflow
gr.renormalize(nel) # Renormalize to the proper # electrons
dens = gr.dens()
weight = gr.weights()
gamma = gr.get_gamma()
npts = len(dens)
if gr.version == 1:
amdens = zeros((2, npts), 'd')
amgamma = zeros((3, npts), 'd')
amdens[0, :] = amdens[1, :] = 0.5 * dens
if gamma is not None:
amgamma[0, :] = amgamma[1, :] = amgamma[2, :] = 0.25 * gamma
elif gr.version == 2:
amdens = gr.density.T
amgamma = gr.gamma.T
fxc, dfxcdna, dfxcdnb, dfxcdgaa, dfxcdgab, dfxcdgbb = XC(
amdens, amgamma, **kwargs)
Exc = dot(weight, fxc)
wva = weight * dfxcdna # Combine w*v in a vector for multiplication by bfs
# First do the part that doesn't depend upon gamma
nbf = gr.get_nbf()
Fxca = zeros((nbf, nbf), 'd')
for i in xrange(nbf):
wva_i = wva * gr.bfgrid[:, i]
for j in xrange(nbf):
Fxca[i, j] = dot(wva_i, gr.bfgrid[:, j])
# Now do the gamma-dependent part.
# Fxc_a += dot(2 dfxcdgaa*graddensa + dfxcdgab*graddensb,grad(chia*chib))
# We can do the dot product as
# sum_grid sum_xyz A[grid,xyz]*B[grid,xyz]
# or a 2d trace product
# Here A contains the dfxcdgaa stuff
# B contains the grad(chia*chib)
# Possible errors: gr.grad() here should be the grad of the b part?
if do_grad_dens:
# A,B are dimensioned (npts,3)
A = transpose(0.5 * transpose(gr.grad()) * (weight *
(2 * dfxcdgaa + dfxcdgab)))
for a in xrange(nbf):
for b in xrange(a + 1):
B = gr.grad_bf_prod(a, b)
Fxca[a, b] += sum(ravel(A * B))
Fxca[b, a] = Fxca[a, b]
if not do_spin_polarized: return Exc, Fxca
wvb = weight * dfxcdnb # Combine w*v in a vector for multiplication by bfs
# First do the part that doesn't depend upon gamma
Fxcb = zeros((nbf, nbf), 'd')
for i in xrange(nbf):
wvb_i = wvb * gr.bfgrid[:, i]
for j in xrange(nbf):
Fxcb[i, j] = dot(wvb_i, gr.bfgrid[:, j])
# Now do the gamma-dependent part.
# Fxc_b += dot(2 dfxcdgbb*graddensb + dfxcdgab*graddensa,grad(chia*chib))
# We can do the dot product as
# sum_grid sum_xyz A[grid,xyz]*B[grid,xyz]
# or a 2d trace product
# Here A contains the dfxcdgaa stuff
# B contains the grad(chia*chib)
# Possible errors: gr.grad() here should be the grad of the b part?
if do_grad_dens:
# A,B are dimensioned (npts,3)
A = transpose(0.5 * transpose(gr.grad()) * (weight *
(2 * dfxcdgbb + dfxcdgab)))
for a in xrange(nbf):
for b in xrange(a + 1):
B = gr.grad_bf_prod(a, b)
Fxcb[a, b] += sum(ravel(A * B))
Fxcb[b, a] = Fxcb[a, b]
return Exc, Fxca, Fxcb
