123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183 |
- #!/usr/bin/env python3
- # -*- coding: UTF-8 -*-
- import numpy as np
- import matplotlib.pyplot as plt
- import os
- import scipy.special.hankel2 as H2n
- c = 299792458.0
- eps_0 = 8.854187817e-12 # F/m
- pi = np.pi
- verbose = 6
- # r of monitor
- r = 146.513e-9
- debug = True
- def read_data(dirname):
- data = {}
- WLs = []
- for r,d,f in os.walk(dirname):
- for fname in f:
- WLs.append(fname)
- for fname in WLs:
- fdata = np.transpose(
- np.genfromtxt(dirname+"/"+fname, delimiter=", ",skip_header=1
- ,dtype=None, encoding = None
- , converters={0: lambda s: complex(s),
- 1: lambda s: complex(s),
- 2: lambda s: complex(s.replace('i', 'j')),
- 3: lambda s: complex(s.replace('i', 'j')),
- 4: lambda s: complex(s.replace('i', 'j')),
- 5: lambda s: complex(s.replace('i', 'j')),
- 6: lambda s: complex(s.replace('i', 'j')),
- 7: lambda s: complex(s.replace('i', 'j')),
- 8: lambda s: complex(s.replace('i', 'j'))
- }
- )
- )
- data[float(fname[2:-4])]=fdata
- if debug: break
- return data
- def find_nearest(array,value):
- idx = (np.abs(array-value)).argmin()
- return array[idx],idx
- def get_WLs_idx(WLs, data):
- dist = 1 #mkm
- mmedia = 1 # vacuum
- shift = 1 # one mesh step
- WLs_idx = []
- for wl in WLs:
- val, idx = find_nearest(data[dist][mmedia][shift][0,:],wl*1e-9)
- WLs_idx.append(idx)
- return WLs_idx
- # def check_field_match(data_in_air, data_in_gold,wl_idx,z_vec,kappa1,kappa2,eps2):
- # z = z_vec[i]*1e-9
- # if verbose > 8: print("z =",z)
- # H1_0 = H1[i]/np.exp(-kappa1[wl_idx]*z)
- # H2_0 = H2[i]/np.exp(-kappa2[wl_idx]*z)
- # E1_0 = E1[i]/np.exp(-kappa1[wl_idx]*z)
- # E2_0 = E2[i]/np.exp(-kappa2[wl_idx]*z)
- # E2_0e = E2[i]/np.exp(-kappa2[wl_idx]*z)*eps2[wl_idx]
- # if verbose > 8:
- # print("H0 air (%5.4g %+5.4gj)"%(np.real(H1_0), np.imag(H1_0)),
- # " from H1 (%5.4g %+5.4gj)"%(np.real(H1[i]), np.imag(H1[i])))
- def analyze(data,wl):
- # print(data[0,:]) # all z values
- #data = "z, dip.power, Ex, Ey, Ez, Hx, Hy, Hz, n_Au"
- # 0, 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 "
- lambd = wl
- omega = 2*pi*c/lambd
- eps_d = complex(1) # air, z>0
- eps_m = data[8,0]**2 # metal, z<0
- dip_power = data[1,0]
- z = data[0,:]
- idx_d = np.nonzero(z>1e-10)
- idx_0 = np.nonzero(np.logical_and(z<=1e-10, z>=-1e-10))
- idx_m = np.nonzero(z<-1e-10)
- z_d = z[idx_d]
- z_0 = z[idx_0]
- z_m = z[idx_m]
- if (not np.array_equal(np.hstack((z_m, z_0, z_d)), z)):
- print("ERROR! loosing z values!")
- raise
-
- Ex = data[2,:]
- Ex_m = data[2,idx_m][0]
- Ey_m = data[3,idx_m][0]
- Ez_m = data[4,idx_m][0]
- Hx_m = data[5,idx_m][0]
- Hy_m = data[6,idx_m][0]
- Hz_m = data[7,idx_m][0]
- E_m = np.array([Ex_m,Ey_m,Ez_m])
- H_m = np.array([Hx_m,Hy_m,Hz_m])
- Ex_d = data[2,idx_d][0]
- Ey_d = data[3,idx_d][0]
- Ez_d = data[4,idx_d][0]
- Hx_d = data[5,idx_d][0]
- Hy_d = data[6,idx_d][0]
- Hz_d = data[7,idx_d][0]
- E_d = np.array([Ex_d,Ey_d,Ez_d])
- H_d = np.array([Hx_d,Hy_d,Hz_d])
- k_0 = omega/c #air
- k_sp = k_0*np.sqrt(eps_d*eps_m/(eps_d+eps_m)) # eq5, supmat
- chi_d = np.sqrt( eps_d*k_0**2 - k_sp**2 ) # desc. after eq6c, supmat
- chi_m = np.sqrt( eps_m*k_0**2 - k_sp**2 ) # desc. after eq6c, supmat
- h_sp_d = np.exp(1j*chi_d*z_d) # eq6a, supmat
- e_sp_x_d = chi_d/(omega*eps_0*eps_d)*np.exp(1j*chi_d*z_d) # eq6b, supmat
- e_sp_z_d = k_sp/(omega*eps_0*eps_d)*np.exp(1j*chi_d*z_d) # eq6c, supmat
- h_sp_m = np.exp(1j*-chi_m*z_m) # eq6a, supmat
- e_sp_x_m = -chi_m/(omega*eps_0*eps_m)*np.exp(1j*-chi_m*z_m) # eq6b, supmat
- e_sp_z_m = k_sp/(omega*eps_0*eps_m)*np.exp(1j*-chi_m*z_m) # eq6c, supmat
- if verbose > 5:
- print("r =",r)
- print(h_sp_m)
- # print("S from full field",np.real(np.cross(E,np.conj(H))))
- # print("H0 air (%5.4g %+5.4gj)"%(np.real(H1_0[wl_idx]), np.imag(H1_0[wl_idx])),
- # " from H1 (%5.4g %+5.4gj)"%(np.real(H1[0][wl_idx]), np.imag(H1[0][wl_idx])))
- # #plasmon_power = 1.0/2.0 * np.real( E1[0] * np.conj(H1[0])) # TODO check minus sign!!
- # plasmon_power = -1.0/2.0 * 2.0*np.pi*R * ( # TODO check minus sign!!
- # np.real( E1_0 * np.conj(H1_0) )
- # / (2.0 * np.real(kappa1))
- # +
- # np.real( E2_0 * np.conj(H1_0) )
- # / (2.0 * np.real(kappa2))
- # )* np.exp( 2.0*np.imag(k_spp)*R ) # TODO check minus sign!!
- # #print(np.abs(plasmon_power/ dip_power))
- # eta0 = plasmon_power[0]/ dip_power[0] *100
- # ppw = plasmon_power[0]
- # print("\n")
- # print(dirname)
- # print("Power: plasmon %4.3g W of dipoles %4.3g W, efficiency %5.3g%% from:"%(ppw, float(np.abs(dip_power[0])),float(np.abs( eta0))), ppw, eta0)
- # plt.plot(lambd*1e9, plasmon_power/ dip_power)
- # plt.ylim(0,0.04)
- # plt.xlim(550,800)
- # #plt.plot(lambd*1e9, np.real(eps2))
- # # plt.plot(lambd*1e9, np.real(k_spp))
- # # plt.plot(lambd*1e9, k_0)
- # #plt.semilogy(lambd*1e9, np.absolute(plasmon_power/ dip_power))
- # # # legend = []
- # # # legend.append(zshift[shift]+"@"+str(WLs[i])+" nm")
- # # # plt.legend(legend)
- # # # #plt.xlabel(r'THz')
- # plt.xlabel(r'$\lambda$, nm')
- # plt.ylabel(r'$P_{spp}/P_{dipole}$',labelpad=-5)
- # #plt.title(' R = '+str(core_r)+' nm')
- # plt.savefig(dirname+"_power_ratio."+file_ext)
- # plt.clf()
- # plt.close()
-
- file_ext="pdf"
- dirname="bigourdan-Au-sub-Cyl-dipole-W.fsp.1D.monitor_1.results"
- def main ():
- data = read_data(dirname)
- for wl in data:
- analyze(data[wl],wl)
- main()
|