qpms/misc/finiterectlat-constant-driv...

381 lines
18 KiB
Python
Executable File

#!/usr/bin/env python3
import math
from qpms.argproc import ArgParser, make_dict_action, sslice, annotate_pdf_metadata
figscale=3
ap = ArgParser(['rectlattice2d_finite', 'single_particle', 'single_lMax_or_vswfset', 'single_omega'])
ap.add_argument("-k", '--wavevector', nargs=2, type=float, required=True, help='"Bloch" vector, modulating phase of the driving', metavar=('KX', 'KY'), default=(0., 0.))
# ap.add_argument("--kpi", action='store_true', help="Indicates that the k vector is given in natural units instead of SI, i.e. the arguments given by -k shall be automatically multiplied by pi / period (given by -p argument)")
ap.add_argument("-o", "--output", type=str, required=False, help='output path (if not provided, will be generated automatically)')
ap.add_argument("-O", "--plot-out", type=str, required=False, help="path to plot output (optional)")
ap.add_argument("-P", "--plot", action='store_true', help="if -p not given, plot to a default path")
ap.add_argument("-g", "--save-gradually", action='store_true', help="saves the partial result after computing each irrep")
ap.add_argument("-S", "--symmetry-adapted", default=None, help="Use a symmetry-adapted basis of a given point group instead of individual spherical harmonics")
ap.add_argument("-d", "--ccd-distance", type=float, default=math.nan, help='Far-field "CCD" distance from the sample')
ap.add_argument("-D", "--ccd-size", type=float, default=math.nan, help='Far-field "CCD" width and heighth')
ap.add_argument("-R", "--ccd-resolution", type=int, default=101, help='Far-field "CCD" resolution')
ap.add_argument("--xslice", default={None:None}, nargs=2,
action=make_dict_action(argtype=sslice, postaction='append', first_is_key=True),
)
ap.add_argument("--yslice", default={None:None}, nargs=2,
action=make_dict_action(argtype=sslice, postaction='append', first_is_key=True),
)
#ap.add_argument("--irrep", type=str, default="none", help="Irrep subspace (irrep index from 0 to 7, irrep label, or 'none' for no irrep decomposition")
a=ap.parse_args()
import logging
logging.basicConfig(format='%(asctime)s %(message)s', level=logging.INFO)
Nx, Ny = a.size
px, py = a.period
particlestr = ("sph" if a.height is None else "cyl") + ("_r%gnm" % (a.radius*1e9))
if a.height is not None: particlestr += "_h%gnm" % (a.height * 1e9)
defaultprefix = "cd_%s_p%gnmx%gnm_%dx%d_m%s_n%s_k_%g_%g_f%geV_%s_micro-%s" % (
particlestr, px*1e9, py*1e9, Nx, Ny, str(a.material), str(a.background), a.wavevector[0], a.wavevector[1], a.eV, ap.bspecstr, "SO3" if a.symmetry_adapted is None else a.symmetry_adapted)
logging.info("Default file prefix: %s" % defaultprefix)
import numpy as np
import qpms
from qpms.cybspec import BaseSpec
from qpms.cytmatrices import CTMatrix, TMatrixGenerator
from qpms.qpms_c import Particle, qpms_library_version
from qpms.cymaterials import EpsMu, EpsMuGenerator, LorentzDrudeModel, lorentz_drude
from qpms.cycommon import DebugFlags, dbgmsg_enable
from qpms import FinitePointGroup, ScatteringSystem, BesselType, eV, hbar
from qpms.symmetries import point_group_info
eh = eV/hbar
# Check slice ranges and generate all corresponding combinations
slicepairs = []
slicelabels = set(a.xslice.keys()) | set(a.yslice.keys())
for label in slicelabels:
rowslices = a.xslice.get(label, None)
colslices = a.yslice.get(label, None)
# TODO check validity of the slices.
if rowslices is None:
rowslices = [slice(None, None, None)]
if colslices is None:
colslices = [slice(None, None, None)]
for rs in rowslices:
for cs in colslices:
slicepairs.append((rs, cs))
def realdipfieldlabels(yp):
if yp == 0: return 'x'
if yp == 1: return 'y'
if yp == 2: return 'z'
raise ValueError
def realdipfields(vecgrid, yp, bspec=BaseSpec(lMax=1)):
if yp == 0 or yp == 1:
im = np.where(bspec.ilist == 6)[0][0]
ip = np.where(bspec.ilist == 14)[0][0]
if yp == 1:
return vecgrid[...,ip] + vecgrid[...,ip]
if yp == 0:
return -1j*(vecgrid[...,im] - vecgrid[...,ip])
if yp == 2:
i0 = np.where(bspec.ilist == 10)[0][0]
return vecgrid[...,i0]
raise ValueError
def float_nicestr(x, tol=1e-5):
x = float(x)
if .5**2 - abs(x) < tol:
return(("-" if x < 0 else '+') + "2^{-2}")
else:
return "%+.3g" % x
def cplx_nicestr(x, tol=1e-5):
x = complex(x)
if x == 0:
return '0'
ret = ""
if x.real:
ret = ret + float_nicestr(x.real, tol)
if x.imag:
ret = ret + float_nicestr(x.imag, tol) + 'i'
if x.real and x.imag:
return '(' + ret + ')'
else:
return ret
def cleanarray(a, atol=1e-10, copy=True):
a = np.array(a, copy=copy)
sieve = abs(a.real) < atol
a[sieve] = 1j * a[sieve].imag
sieve = abs(a.imag) < atol
a[sieve] = a[sieve].real
return a
def nicerot(a, atol=1e-10, copy=True): #gives array a "nice" phase
a = np.array(a, copy=copy)
i = np.argmax(abs(a))
a = a / a[i] * abs(a[i])
return a
dbgmsg_enable(DebugFlags.INTEGRATION)
#Particle positions
orig_x = (np.arange(Nx/2) + (0 if (Nx % 2) else .5)) * px
orig_y = (np.arange(Ny/2) + (0 if (Ny % 2) else .5)) * py
orig_xy = np.stack(np.meshgrid(orig_x, orig_y), axis = -1)
omega = ap.omega
bspec = ap.bspec
medium = EpsMuGenerator(ap.background_epsmu)
particles= [Particle(orig_xy[i], ap.tmgen, bspec) for i in np.ndindex(orig_xy.shape[:-1])]
sym = FinitePointGroup(point_group_info['D2h'])
ss, ssw = ScatteringSystem.create(particles=particles, medium=medium, omega=omega, sym=sym)
wavenumber = ap.background_epsmu.k(omega) # Currently, ScatteringSystem does not "remember" frequency nor wavenumber
# Mapping between ss particles and grid positions
positions = ss.positions
xpositions = np.unique(positions[:,0])
assert(len(xpositions) == Nx)
ypositions = np.unique(positions[:,1])
assert(len(ypositions == Ny))
# particle positions as integer indices
posmap = np.empty((positions.shape[0],2), dtype=int)
invposmap = np.empty((Nx, Ny), dtype=int)
for i, pos in enumerate(positions):
posmap[i,0] = np.searchsorted(xpositions, positions[i,0])
posmap[i,1] = np.searchsorted(ypositions, positions[i,1])
invposmap[posmap[i,0], posmap[i, 1]] = i
def fullvec2grid(fullvec, swapxy=False):
arr = np.empty((Nx,Ny,nelem), dtype=complex)
for pi, offset in enumerate(ss.fullvec_poffsets):
ix, iy = posmap[pi]
arr[ix, iy] = fullvec[offset:offset+nelem]
return np.swapaxes(arr, 0, 1) if swapxy else arr
outfile_tmp = defaultprefix + ".tmp" if a.output is None else a.output + ".tmp"
nelem = len(bspec)
phases = np.exp(1j*np.dot(ss.positions[:,:2], np.array(a.wavevector)))
driving_full = np.zeros((nelem, ss.fecv_size),dtype=complex)
if a.symmetry_adapted is not None:
ss1, ssw1 = ScatteringSystem.create(particles=[Particle((0,0,0), ap.tmgen, bspec)], medium=medium, omega=omega,
sym=FinitePointGroup(point_group_info[a.symmetry_adapted]))
fvcs1 = np.empty((nelem, nelem), dtype=complex)
y = 0
iris1 = []
for iri1 in range(ss1.nirreps):
for j in range(ss1.saecv_sizes[iri1]):
pvc1 = np.zeros((ss1.saecv_sizes[iri1],), dtype=complex)
pvc1[j] = 1
fvcs1[y] = ss1.unpack_vector(pvc1, iri1)
fvcs1[y] = cleanarray(nicerot(fvcs1[y], copy=False), copy=False)
driving_full[y] = (phases[:, None] * fvcs1[y][None,:]).flatten()
y += 1
iris1.append(iri1)
iris1 = np.array(iris1)
else:
for y in range(nelem):
driving_full[y,y::nelem] = phases
# Apply the driving on the specified slices only
nsp = len(slicepairs)
driving_full_sliced = np.zeros((nsp,) + driving_full.shape, dtype=complex)
p1range = np.arange(nelem)
for spi in range(nsp):
xs, ys = slicepairs[spi]
driven_pi = invposmap[xs, ys].flatten()
driven_y = ((driven_pi * nelem)[:,None] + p1range[None,:]).flatten()
driving_full_sliced[spi][:, driven_y] = driving_full[:, driven_y]
scattered_full = np.zeros((nsp, nelem, ss.fecv_size),dtype=complex)
scattered_ir = [None for iri in range(ss.nirreps)]
ir_contained = np.ones((nsp, nelem, ss.nirreps), dtype=bool)
for iri in range(ss.nirreps):
if ss.saecv_sizes[iri] == 0:
logging.info('irrep %d/%d has an empty VSWF set, skipping')
continue
logging.info("processing irrep %d/%d" % (iri, ss.nirreps))
LU = None # to trigger garbage collection before the next call
translation_matrix = None
LU = ssw.scatter_solver(iri)
logging.info("LU solver created")
#translation_matrix = ss.translation_matrix_packed(wavenumber, iri, BesselType.REGULAR) + np.eye(ss.saecv_sizes[iri])
#logging.info("auxillary translation matrix created")
scattered_ir[iri] = np.zeros((nsp, nelem, ss.saecv_sizes[iri]), dtype=complex)
scattered_ir_unpacked = np.zeros((nsp, nelem, ss.fecv_size), dtype=complex)
for spi in range(nsp):
for y in range(nelem):
ã = driving_full_sliced[spi,y]
ãi = cleanarray(ss.pack_vector(ã, iri), copy=False)
if np.all(ãi == 0):
ir_contained[spi, y, iri] = False
continue
= ssw.apply_Tmatrices_full(ã)
Tãi = ss.pack_vector(, iri)
fi = LU(Tãi)
scattered_ir[iri][spi, y] = fi
scattered_ir_unpacked[spi, y] = ss.unpack_vector(fi, iri)
scattered_full[spi, y] += scattered_ir_unpacked[spi, y]
if a.save_gradually:
iriout = outfile_tmp + ".%d" % iri
np.savez(iriout, iri=iri, meta={**vars(a), 'qpms_version' : qpms.__version__()},
omega=omega, wavenumber=wavenumber, nelem=nelem, wavevector=np.array(a.wavevector), phases=phases,
positions = ss.positions[:,:2],
scattered_ir_packed = scattered_ir[iri],
scattered_ir_full = scattered_ir_unpacked,
)
logging.info("partial results saved to %s"%iriout)
t, l, m = bspec.tlm()
if not math.isnan(a.ccd_distance):
logging.info("Computing the far fields")
if math.isnan(a.ccd_size):
a.ccd_size = (50 * a.ccd_distance / (max(Nx*px, Ny*py) *ssw.wavenumber.real))
ccd_size = a.ccd_size
ccd_x = np.linspace(-ccd_size/2, ccd_size/2, a.ccd_resolution)
ccd_y = np.linspace(-ccd_size/2, ccd_size/2, a.ccd_resolution)
ccd_grid = np.meshgrid(ccd_x, ccd_y, (a.ccd_distance,), indexing='ij')
ccd_points = np.swapaxes(np.stack(ccd_grid, axis=-1).squeeze(axis=-2), 0,1) # First axis is y, second is x, because of imshow...
ccd_fields = np.empty((nsp, nelem,) + ccd_points.shape, dtype=complex)
for spi in range(nsp):
for y in range(nelem):
ccd_fields[spi, y] = ssw.scattered_E(scattered_full[spi, y], ccd_points, btyp=BesselType.HANKEL_PLUS)
logging.info("Far fields done")
outfile = defaultprefix + ".npz" if a.output is None else a.output
np.savez(outfile, meta={**vars(a), 'qpms_version' : qpms.__version__()},
omega=omega, wavenumber=wavenumber, nelem=nelem, wavevector=np.array(a.wavevector), phases=phases,
positions = ss.positions[:,:2],
scattered_ir_packed = np.array(scattered_ir, dtype=np.object),
scattered_full = scattered_full,
ir_contained = ir_contained,
t=t, l=l, m=m,
iris1 = iris1 if (a.symmetry_adapted is not None) else None,
irnames1 = ss1.irrep_names if (a.symmetry_adapted is not None) else None,
fvcs1 = fvcs1 if (a.symmetry_adapted is not None) else None,
#ccd_size = ccd_size if not math.isnan(a.ccd_distance) else None,
ccd_points = ccd_points if not math.isnan(a.ccd_distance) else None,
ccd_fields = ccd_fields if not math.isnan(a.ccd_distance) else None,
fullvec_poffsets = ss.fullvec_poffsets,
)
logging.info("Saved to %s" % outfile)
if a.plot or (a.plot_out is not None):
import matplotlib
matplotlib.use('pdf')
from matplotlib import pyplot as plt, cm
from matplotlib.backends.backend_pdf import PdfPages
t, l, m = bspec.tlm()
phasecm = cm.twilight
pmcm = cm.bwr
abscm = cm.plasma
plotfile = defaultprefix + ".pdf" if a.plot_out is None else a.plot_out
pp = PdfPages(plotfile)
for spi in range(nsp):
fig, axes = plt.subplots(nelem, 12 if math.isnan(a.ccd_distance) else 16, figsize=(figscale*(12 if math.isnan(a.ccd_distance) else 16), figscale*nelem))
for yp in range(0,3): # TODO xy-dipoles instead?
axes[0,4*yp+0].set_title("abs / (E,1,%s)" % realdipfieldlabels(yp))
axes[0,4*yp+1].set_title("arg / (E,1,%s)" % realdipfieldlabels(yp))
axes[0,4*yp+2].set_title("Fabs / (E,1,%s)" % realdipfieldlabels(yp))
axes[0,4*yp+3].set_title("Farg / (E,1,%s)" % realdipfieldlabels(yp))
if not math.isnan(a.ccd_distance):
#axes[0,12].set_title("$E_{xy}$ @ $z = %g; \phi$" % a.ccd_distance)
#axes[0,13].set_title("$E_{xy}$ @ $z = %g; \phi + \pi/2$" % a.ccd_distance)
axes[0,12].set_title("$|E_{x}|^2$ @ $z = %g\,\mathrm{m}$" % a.ccd_distance)
axes[0,13].set_title("$|E_{y}|^2$ @ $z = %g\,\mathrm{m}$" % a.ccd_distance)
axes[0,14].set_title("$|E_x + E_y|^2$ @ $z = %g\,\mathrm{m}$" % a.ccd_distance)
axes[0,15].set_title("$|E_{z}|^2$ @ $z = %g\,\mathrm{m}$" % a.ccd_distance)
for gg in range(12,16):
axes[-1,gg].set_xlabel("$x/\mathrm{m}$")
for y in range(nelem):
fulvec = scattered_full[spi,y]
if a.symmetry_adapted is not None:
driving_nonzero_y = [j for j in range(nelem) if abs(fvcs1[y,j]) > 1e-5]
driving_descr = ss1.irrep_names[iris1[y]]+'\n'+', '.join(('$'+cplx_nicestr(fvcs1[y,j])+'$' +
"(%s,%d,%+d)" % (("E" if t[j] == 2 else "M"), l[j], m[j]) for j in
driving_nonzero_y)) # TODO shorten the complex number precision
else:
driving_descr = "%s,%d,%+d"%('E' if t[y]==2 else 'M', l[y], m[y],)
axes[y,0].set_ylabel(driving_descr)
axes[y,-1].yaxis.set_label_position("right")
axes[y,-1].set_ylabel("$y/\mathrm{m}$\n"+driving_descr)
vecgrid = fullvec2grid(fulvec, swapxy=True)
vecgrid_ff = np.fft.fftshift(np.fft.fft2(vecgrid, axes=(0,1)),axes=(0,1))
lemax = np.amax(abs(vecgrid))
for yp in range(0,3):
try:
if(np.amax(abs(realdipfields(vecgrid,yp,bspec))) > lemax*1e-5):
axes[y,yp*4].imshow(abs(realdipfields(vecgrid,yp,bspec)), vmin=0, interpolation='none')
axes[y,yp*4].text(0.5, 0.5, '%g' % np.amax(abs(realdipfields(vecgrid,yp,bspec))), horizontalalignment='center', verticalalignment='center', transform=axes[y,yp*4].transAxes)
axes[y,yp*4+1].imshow(np.angle(realdipfields(vecgrid,yp,bspec)), vmin=-np.pi, vmax=np.pi, cmap=phasecm, interpolation='none')
axes[y,yp*4+2].imshow(abs(realdipfields(vecgrid_ff,yp,bspec)), vmin=0, interpolation='none')
axes[y,yp*4+3].imshow(np.angle(realdipfields(vecgrid_ff,yp,bspec)), vmin=-np.pi, vmax=np.pi, cmap=phasecm, interpolation='none')
else:
for c in range(0,4):
axes[y,yp*4+c].tick_params(bottom=False, left=False, labelbottom=False, labelleft=False)
except (KeyError, IndexError) as e:
logging.info("skipping dipole plot #%d for y=%d, spi=%d (dipole probably not included in the VSWF set) %s" %
(yp, y, spi, e))
for c in range(0,4):
axx= axes[y,yp*4+c]
axx.tick_params(bottom=False, left=False, labelbottom=False, labelleft=False)
axx.text(0.5, 0.5, 'skipped', horizontalalignment='center',verticalalignment='center', transform=axx.transAxes)
if not math.isnan(a.ccd_distance):
fxye=(-ccd_size/2, ccd_size/2, -ccd_size/2, ccd_size/2)
e2vmax = np.amax(np.linalg.norm(ccd_fields[spi,y], axis=-1)**2)
xint = abs(ccd_fields[spi,y,...,0])**2
yint = abs(ccd_fields[spi,y,...,1])**2
xyint = abs(ccd_fields[spi,y,...,0] + ccd_fields[spi,y,...,1])**2
zint = abs(ccd_fields[spi,y,...,2])**2
xintmax = np.amax(xint)
yintmax = np.amax(yint)
zintmax = np.amax(zint)
xyintmax = np.amax(xyint)
axes[y, 12].imshow(xint, origin="lower", extent=fxye, cmap=abscm, interpolation='none')
axes[y, 13].imshow(yint, origin="lower", extent=fxye, cmap=abscm, interpolation='none')
axes[y, 14].imshow(xyint, origin="lower", extent=fxye, cmap=abscm, interpolation='none')
axes[y, 15].imshow(zint, origin='lower', extent=fxye, cmap=abscm, interpolation='none')
axes[y, 12].text(0.5, 0.5, '%g\n%g' % (xintmax,xintmax/e2vmax),
horizontalalignment='center', verticalalignment='center', transform=axes[y,12].transAxes)
axes[y, 13].text(0.5, 0.5, '%g\n%g' % (yintmax,yintmax/e2vmax),
horizontalalignment='center', verticalalignment='center', transform=axes[y,13].transAxes)
axes[y, 14].text(0.5, 0.5, '%g\n%g' % (xyintmax,xyintmax/e2vmax),
horizontalalignment='center', verticalalignment='center', transform=axes[y,14].transAxes)
axes[y, 15].text(0.5, 0.5, '%g\n%g' % (zintmax,zintmax/e2vmax),
horizontalalignment='center', verticalalignment='center', transform=axes[y,15].transAxes)
for gg in range(12,16):
axes[y,gg].yaxis.tick_right()
for gg in range(12,15):
axes[y,gg].yaxis.set_major_formatter(plt.NullFormatter())
fig.text(0, 0, str(slicepairs[spi]), horizontalalignment='left', verticalalignment='bottom')
pp.savefig()
annotate_pdf_metadata(pp, scriptname="finiterectlat-constant-driving.py")
pp.close()
exit(0)