Dear developers, dear colleagues,
Kwant is a really excellent package and I appreciate what you have done to
make the numerical simulation easier.
I have been using Kwant for a while. Recently I am simulating the
disordered quantum anomalous Hall insulators. The disorder average is
time-consuming. In the simulation the Fermi energy of the central scattering
region is fixed, and I know that extracting the conductance involves a step of
computing modes in the leads from
https://www.mail-archive.com/kwant-discuss@kwant-project.org/msg00718.html,
thus it came to me that whether I can solve the modes in the leads and store
them before, and then I just invoke instead of solving them for every disorder
configuration. I have tried to realized a MatLab implementation of
Landauer-Buttiker formula based on the recursive Green's function, in which I
solved the self energy and line width function of leads for once and performing
the disorder average with them invoked and it works fine. I know that the
implementation in Kwant is some kind of different, and I was wondering if there
was something optimized for simulating disordered system.
Below is my code:
import kwant
import numpy as np
import time as ti
from mpi4py import MPI
comm = MPI.COMM_WORLD
k = comm.Get_rank()
nprocessor = comm.size
import warnings
warnings.filterwarnings('ignore')
# Pauli matrices
s0 = np.eye(2)
sx = np.array([[0,1], [1,0]])
sy = np.array([[0, -1j],[1j, 0]])
sz = np.array([[1,0],[0,-1]])
# Parameters
A = 364.5 # meV nm
B = -686 # meV nm^2
C = 0 # meV
D = -512 # meV nm^2
M = -10 # meV
L = 256
# Building the system
def qwz(a=1, L=L, W=L):
hamiltonian_syst = """
(V(x, y) + C-D*(k_x**2+k_y**2))* identity(2)
+ (M-B*(k_x**2+k_y**2)) * sigma_z
+ A*k_x*sigma_x
+ A*k_y*sigma_y
"""
hamiltonian_lead = """
(C-D*(k_x**2+k_y**2))* identity(2)
+ (M-B*(k_x**2+k_y**2)) * sigma_z
+ A*k_x*sigma_x
+ A*k_y*sigma_y
"""
template_syst = kwant.continuum.discretize(hamiltonian_syst, grid=a)
template_lead = kwant.continuum.discretize(hamiltonian_lead, grid=a)
def shape(site):
(x, y) = site.pos
return (0 <= y < W and 0 <= x < L)
def lead_shape(site):
(x, y) = site.pos
return (0 <= y < W)
syst = kwant.Builder()
syst.fill(template_syst, shape, (0, 0))
lead = kwant.Builder(kwant.TranslationalSymmetry([-a, 0]))
lead.fill(template_lead, lead_shape, (0, 0))
syst.attach_lead(lead)
syst.attach_lead(lead.reversed())
# kwant.plot(syst)
syst = syst.finalized()
return syst
def get_conductance(w, Ef, Nc): # w is the disorder strength, Ef is the Fermi
energy, and Nc is the cycling times
def disorder(x, y): # onsite disorder
return np.random.uniform(-w/2, w/2)
params = dict(A = A, B = B, C = C, D = D, M = M, Ef = Ef, V = disorder)
syst = qwz()
data = []
for i in range(Nc):
smatrix = kwant.smatrix(syst, Ef, params = params)
data.append(smatrix.transmission(1, 0))
G_av = np.mean(data)
G_std = np.std(data)
return G_av, G_std
def main():
t1 = ti.time()
syst = qwz()
Ef = 100 # meV
Nc = 500 # cycle
w = k * 20 # meV
G_av, G_std = get_conductance(w, Ef, Nc)
# pyplot.plot(energies, data)
# pyplot.xlabel("energy [t]")
# pyplot.ylabel("conductance [e^2/h]")
# pyplot.show()
# pyplot.savefig('conductance' +str(k)+ '.png')
np.savez('cond('+str(L)+')'+str(k)+ '.npz', x = w, y = G_av, ye = G_std)
# print('Disorder strength:', w, 'Average conductance:', G_av, 'Conductance
fluctuation:', G_std)
t2 = ti.time()
print('time cost:', t2-t1, 's')
if __name__ == '__main__':
main()
In which I simulated a system of 256*256 atoms and averaged for 500 samples. It
will take about 20000 seconds with a dual E5 2696 v3 server with 50 threads
working parallely at about 2.7 GHz. I know that part of the developers
contributed to the termed Topological Anderson Insulator, and I was wondering
if there's some way to improve the efficiency of my code?
Thank you for taking your time!
Yours sincerely,
Shihao, A PhD candidate in cond-mat theory
