def __init__(
self,
model,
angle,
L12=2859,
footprint=60,
L2S=120,
dtheta=3.3, # angular resolution
lo_wavelength=2.8,
hi_wavelength=18,
dlambda=3.3,
rebin=2):
self.model = model
# turn off resolution smearing
self.model.dq = 0
self.bkg = model.bkg.value
self.angle = angle
# the fractional width of a square wavelength resolution
self.dlambda = dlambda / 100.
self.rebin = rebin / 100.
self.wavelength_bins = calculate_wavelength_bins(
lo_wavelength, hi_wavelength, rebin)
# nominal Q values
bin_centre = 0.5 * (self.wavelength_bins[1:] +
self.wavelength_bins[:-1])
self.q = general.q(angle, bin_centre)
# keep a tally of the direct and reflected beam
self.direct_beam = np.zeros((self.wavelength_bins.size - 1))
self.reflected_beam = np.zeros((self.wavelength_bins.size - 1))
# wavelength generator
a = PN('PLP0000711.nx.hdf')
q, i, di = a.process(normalise=False,
normalise_bins=False,
rebin_percent=0,
lo_wavelength=lo_wavelength,
hi_wavelength=hi_wavelength)
q = q.squeeze()
i = i.squeeze()
self.spectrum_dist = SpectrumDist(q, i)
# angular resolution generator, based on a trapezoidal distribution
# The slit settings are the optimised set typically used in an
# experiment
self.dtheta = dtheta / 100.
self.footprint = footprint
s1, s2 = general.slit_optimiser(footprint,
self.dtheta,
angle=angle,
L2S=L2S,
L12=L12,
verbose=False)
div, alpha, beta = general.div(s1, s2, L12=L12)
self.angular_dist = trapz(c=(alpha - beta) / 2. / alpha,
d=(alpha + beta) / 2. / alpha,
loc=-alpha,
scale=2 * alpha)
def run(self, samples):
"""
Sample the beam.
2400000 samples roughly corresponds to 1200 sec of *PLATYPUS* using
dlambda=3.3 and dtheta=3.3 at angle=0.65.
150000000 samples roughly corresponds to 3600 sec of *PLATYPUS* using
dlambda=3.3 and dtheta=3.3 at angle=3.0.
(The sample number <--> actual acquisition time correspondence has
not been checked fully)
Parameters
----------
samples: int
How many samples to run.
"""
# generate neutrons of different angular divergence
angles = self.angular_dist.rvs(samples) + self.angle
# generate neutrons of various wavelengths
wavelengths = self.spectrum_dist.rvs(size=samples)
# calculate Q
q = general.q(angles, wavelengths)
# calculate reflectivities for a neutron of a given Q.
# resolution smearing is taken care of elsewhere.
r = self.model(q, x_err=0.)
# accept or reject neutrons based on the reflectivity of
# sample at a given Q.
criterion = np.random.random(size=samples)
accepted = criterion < r
# implement wavelength smearing from choppers
# factor of 0.68 is used to convert from FWHM-Gaussian to
# full-width-Uniform
noise = np.random.random(size=samples) - 0.5
jittered_wavelengths = wavelengths * (1 + self.dlambda / 0.68 * noise)
# update direct and reflected beam counts. Rebin smearing
# is taken into account due to the finite size of the wavelength
# bins.
hist = np.histogram(jittered_wavelengths, self.wavelength_bins)
self.direct_beam += hist[0]
hist = np.histogram(jittered_wavelengths[accepted],
self.wavelength_bins)
self.reflected_beam += hist[0]
def _reduce_single_angle(self, scale=1):
"""
Reduce a single angle.
"""
n_spectra = self.reflected_beam.n_spectra
n_tpixels = np.size(self.reflected_beam.m_topandtail, 1)
n_ypixels = np.size(self.reflected_beam.m_topandtail, 2)
# calculate omega and two_theta depending on the mode.
mode = self.reflected_beam.mode
# we'll need the wavelengths to calculate Q.
wavelengths = self.reflected_beam.m_lambda
m_twotheta = np.zeros((n_spectra, n_tpixels, n_ypixels))
detector_z_difference = (self.reflected_beam.detector_z -
self.direct_beam.detector_z)
beampos_z_difference = (self.reflected_beam.m_beampos -
self.direct_beam.m_beampos)
Y_PIXEL_SPACING = self.reflected_beam.cat.y_pixels_per_mm[0]
total_z_deflection = (detector_z_difference +
beampos_z_difference * Y_PIXEL_SPACING)
if mode in ['FOC', 'POL', 'POLANAL', 'MT']:
# omega_nom.shape = (N, )
omega_nom = np.degrees(
np.arctan(total_z_deflection / self.reflected_beam.detector_y)
/ 2.)
'''
Wavelength specific angle of incidence correction
This involves:
1) working out the trajectory of the neutrons through the
collimation system.
2) where those neutrons intersect the sample.
3) working out the elevation of the neutrons when they hit the
sample.
4) correcting the angle of incidence.
'''
speeds = general.wavelength_velocity(wavelengths)
collimation_distance = self.reflected_beam.cat.collimation_distance
s2_sample_distance = (self.reflected_beam.cat.sample_distance -
self.reflected_beam.cat.slit2_distance)
# work out the trajectories of the neutrons for them to pass
# through the collimation system.
trajectories = find_trajectory(collimation_distance / 1000., 0,
speeds)
# work out where the beam hits the sample
res = parabola_line_intersection_point(s2_sample_distance / 1000,
0, trajectories, speeds,
omega_nom[:, np.newaxis])
intersect_x, intersect_y, x_prime, elevation = res
# correct the angle of incidence with a wavelength dependent
# elevation.
omega_corrected = omega_nom[:, np.newaxis] - elevation
m_twotheta += np.arange(n_ypixels * 1.)[np.newaxis, np.newaxis, :]
m_twotheta -= self.direct_beam.m_beampos[:, np.newaxis, np.newaxis]
m_twotheta *= Y_PIXEL_SPACING
m_twotheta += detector_z_difference
m_twotheta /= (self.reflected_beam.detector_y[:, np.newaxis,
np.newaxis])
m_twotheta = np.arctan(m_twotheta)
m_twotheta = np.degrees(m_twotheta)
# you may be reflecting upside down, reverse the sign.
upside_down = np.sign(omega_corrected[:, 0])
m_twotheta *= upside_down[:, np.newaxis, np.newaxis]
omega_corrected *= upside_down[:, np.newaxis]
elif mode in ['SB', 'DB']:
# the angle of incidence is half the two theta of the reflected
# beam
omega = np.arctan(
total_z_deflection / self.reflected_beam.detector_y) / 2.
# work out two theta for each of the detector pixels
m_twotheta += np.arange(n_ypixels * 1.)[np.newaxis, np.newaxis, :]
m_twotheta -= self.direct_beam.m_beampos[:, np.newaxis, np.newaxis]
m_twotheta += detector_z_difference
m_twotheta -= (
self.reflected_beam.detector_y[:, np.newaxis, np.newaxis] *
np.tan(omega[:, np.newaxis, np.newaxis]))
m_twotheta /= (self.reflected_beam.detector_y[:, np.newaxis,
np.newaxis])
m_twotheta = np.arctan(m_twotheta)
m_twotheta += omega[:, np.newaxis, np.newaxis]
# still in radians at this point
# add an extra dimension, because omega_corrected needs to be the
# angle of incidence for each wavelength. I.e. should be
# broadcastable to (N, T)
omega_corrected = np.degrees(omega)[:, np.newaxis]
m_twotheta = np.degrees(m_twotheta)
'''
--Specular Reflectivity--
Use the (constant wavelength) spectra that have already been integrated
over 2theta (in processnexus) to calculate the specular reflectivity.
Beware: this is because m_topandtail has already been divided through
by monitor counts and error propagated (at the end of processnexus).
Thus, the 2theta pixels are correlated to some degree. If we use the 2D
plot to calculate reflectivity
(sum {Iref_{2theta, lambda}}/I_direct_{lambda}) then the error bars in
the reflectivity turn out much larger than they should be.
'''
ydata, ydata_sd = EP.EPdiv(self.reflected_beam.m_spec,
self.reflected_beam.m_spec_sd,
self.direct_beam.m_spec,
self.direct_beam.m_spec_sd)
# calculate the 1D Qz values.
xdata = general.q(omega_corrected, wavelengths)
xdata_sd = (self.reflected_beam.m_lambda_fwhm /
self.reflected_beam.m_lambda)**2
xdata_sd += (self.reflected_beam.domega[:, np.newaxis] /
omega_corrected)**2
xdata_sd = np.sqrt(xdata_sd) * xdata
'''
---Offspecular reflectivity---
normalise the counts in the reflected beam by the direct beam
spectrum this gives a reflectivity. Also propagate the errors,
leaving the fractional variance (dr/r)^2.
--Note-- that adjacent y-pixels (same wavelength) are correlated in
this treatment, so you can't just sum over them.
i.e. (c_0 / d) + ... + c_n / d) != (c_0 + ... + c_n) / d
'''
m_ref, m_ref_sd = EP.EPdiv(
self.reflected_beam.m_topandtail,
self.reflected_beam.m_topandtail_sd,
self.direct_beam.m_spec[:, :, np.newaxis],
self.direct_beam.m_spec_sd[:, :, np.newaxis])
# you may have had divide by zero's.
m_ref = np.where(np.isinf(m_ref), 0, m_ref)
m_ref_sd = np.where(np.isinf(m_ref_sd), 0, m_ref_sd)
# calculate the Q values for the detector pixels. Each pixel has
# different 2theta and different wavelength, ASSUME that they have the
# same angle of incidence
qx, qy, qz = general.q2(omega_corrected[:, :, np.newaxis], m_twotheta,
0, wavelengths[:, :, np.newaxis])
reduction = {}
reduction['x'] = self.x = xdata
reduction['x_err'] = self.x_err = xdata_sd
reduction['y'] = self.y = ydata / scale
reduction['y_err'] = self.y_err = ydata_sd / scale
reduction['omega'] = omega_corrected
reduction['m_twotheta'] = m_twotheta
reduction['m_ref'] = self.m_ref = m_ref
reduction['m_ref_err'] = self.m_ref_err = m_ref_sd
reduction['qz'] = self.m_qz = qz
reduction['qx'] = self.m_qx = qx
reduction['nspectra'] = self.n_spectra = n_spectra
reduction['start_time'] = self.reflected_beam.start_time
reduction['datafile_number'] = self.datafile_number = (
self.reflected_beam.datafile_number)
fnames = []
datasets = []
datafilename = self.reflected_beam.datafilename
datafilename = os.path.basename(datafilename.split('.nx.hdf')[0])
for i in range(n_spectra):
data_tup = self.data(scanpoint=i)
datasets.append(ReflectDataset(data_tup))
if self.save:
for i, dataset in enumerate(datasets):
fname = '{0}_{1}.dat'.format(datafilename, i)
fnames.append(fname)
with open(fname, 'wb') as f:
dataset.save(f)
fname = '{0}_{1}.xml'.format(datafilename, i)
with open(fname, 'wb') as f:
dataset.save_xml(f, start_time=reduction['start_time'][i])
reduction['fname'] = fnames
return datasets, deepcopy(reduction)
def _reduce_single_angle(self, scale=1):
"""
Reduce a single angle.
"""
n_spectra = self.reflected_beam.n_spectra
n_tpixels = np.size(self.reflected_beam.m_topandtail, 1)
n_ypixels = np.size(self.reflected_beam.m_topandtail, 2)
# calculate omega and two_theta depending on the mode.
mode = self.reflected_beam.mode
# we'll need the wavelengths to calculate Q.
wavelengths = self.reflected_beam.m_lambda
m_twotheta = np.zeros((n_spectra, n_tpixels, n_ypixels))
if mode in ['FOC', 'POL', 'POLANAL', 'MT']:
detector_z_difference = (self.reflected_beam.detector_z -
self.direct_beam.detector_z)
beampos_z_difference = (self.reflected_beam.m_beampos
- self.direct_beam.m_beampos)
total_z_deflection = (detector_z_difference
+ beampos_z_difference * Y_PIXEL_SPACING)
# omega_nom.shape = (N, )
omega_nom = np.degrees(np.arctan(total_z_deflection
/ self.reflected_beam.detector_y) / 2.)
'''
Wavelength specific angle of incidence correction
This involves:
1) working out the trajectory of the neutrons through the
collimation system.
2) where those neutrons intersect the sample.
3) working out the elevation of the neutrons when they hit the
sample.
4) correcting the angle of incidence.
'''
speeds = general.wavelength_velocity(wavelengths)
collimation_distance = self.reflected_beam.cat.collimation_distance
s2_sample_distance = (self.reflected_beam.cat.sample_distance
- self.reflected_beam.cat.slit2_distance)
# work out the trajectories of the neutrons for them to pass
# through the collimation system.
trajectories = pm.find_trajectory(collimation_distance / 1000.,
0, speeds)
# work out where the beam hits the sample
res = pm.parabola_line_intersection_point(s2_sample_distance / 1000,
0,
trajectories,
speeds,
omega_nom[:, np.newaxis])
intersect_x, intersect_y, x_prime, elevation = res
# correct the angle of incidence with a wavelength dependent
# elevation.
omega_corrected = omega_nom[:, np.newaxis] - elevation
elif mode == 'SB' or mode == 'DB':
omega = self.reflected_beam.M_beampos + self.reflected_beam.detectorZ[:, np.newaxis]
omega -= self.direct_beam.M_beampos + self.direct_beam.detectorZ
omega /= 2 * self.reflected_beam.detectorY[:, np.newaxis, np.newaxis]
omega = np.arctan(omega)
m_twotheta += np.arange(n_ypixels * 1.)[np.newaxis, np.newaxis, :] * Y_PIXEL_SPACING
m_twotheta += self.reflected_beam.detectorZ[:, np.newaxis, np.newaxis]
m_twotheta -= self.direct_beam.M_beampos[:, :, np.newaxis] + self.direct_beam.detectorZ
m_twotheta -= self.reflected_beam.detectorY[:, np.newaxis, np.newaxis] * np.tan(omega[:, :, np.newaxis])
m_twotheta /= self.reflected_beam.detectorY[:, np.newaxis, np.newaxis]
m_twotheta = np.arctan(m_twotheta)
m_twotheta += omega[:, :, np.newaxis]
'''
--Specular Reflectivity--
Use the (constant wavelength) spectra that have already been integrated
over 2theta (in processnexus) to calculate the specular reflectivity.
Beware: this is because m_topandtail has already been divided through
by monitor counts and error propagated (at the end of processnexus).
Thus, the 2theta pixels are correlated to some degree. If we use the 2D
plot to calculate reflectivity
(sum {Iref_{2theta, lambda}}/I_direct_{lambda}) then the error bars in
the reflectivity turn out much larger than they should be.
'''
ydata, ydata_sd = EP.EPdiv(self.reflected_beam.m_spec,
self.reflected_beam.m_spec_sd,
self.direct_beam.m_spec,
self.direct_beam.m_spec_sd)
# calculate the 1D Qz values.
xdata = general.q(omega_corrected, wavelengths)
xdata_sd = (self.reflected_beam.m_lambda_fwhm
/ self.reflected_beam.m_lambda) ** 2
xdata_sd += (self.reflected_beam.domega[:, np.newaxis]
/ omega_corrected) ** 2
xdata_sd = np.sqrt(xdata_sd) * xdata
'''
---Offspecular reflectivity---
normalise the counts in the reflected beam by the direct beam
spectrum this gives a reflectivity. Also propagate the errors,
leaving the fractional variance (dr/r)^2.
--Note-- that adjacent y-pixels (same wavelength) are correlated in this
treatment, so you can't just sum over them.
i.e. (c_0 / d) + ... + c_n / d) != (c_0 + ... + c_n) / d
'''
m_ref, m_ref_sd = EP.EPdiv(self.reflected_beam.m_topandtail,
self.reflected_beam.m_topandtail_sd,
self.direct_beam.m_spec[:, :, np.newaxis],
self.direct_beam.m_spec_sd[:, :, np.newaxis])
# you may have had divide by zero's.
m_ref = np.where(np.isinf(m_ref), 0, m_ref)
m_ref_sd = np.where(np.isinf(m_ref_sd), 0, m_ref_sd)
# calculate the Q values for the detector pixels. Each pixel has
# different 2theta and different wavelength, ASSUME that they have the
# same angle of incidence
qx, qy, qz = general.q2(omega_corrected[:, :, np.newaxis],
m_twotheta,
0,
wavelengths[:, :, np.newaxis])
reduction = {}
reduction['xdata'] = self.xdata = xdata
reduction['xdata_sd'] = self.xdata_sd = xdata_sd
reduction['ydata'] = self.ydata = ydata
reduction['ydata_sd'] = self.ydata_sd = ydata_sd
reduction['m_ref'] = self.m_ref = m_ref
reduction['m_ref_sd'] = self.m_ref_sd = m_ref_sd
reduction['qz'] = self.m_qz = qz
reduction['qy'] = self.m_qy = qy
reduction['nspectra'] = self.n_spectra = n_spectra
reduction['datafile_number'] = self.datafile_number = (
self.reflected_beam.datafile_number)
fnames = []
if self.save:
for i in range(n_spectra):
data_tup = self.data(scanpoint=i)
dataset = ReflectDataset(data_tup)
fname = 'PLP{0:07d}_{1}.dat'.format(self.datafile_number, i)
fnames.append(fname)
with open(fname, 'wb') as f:
dataset.save(f)
fname = 'PLP{0:07d}_{1}.xml'.format(self.datafile_number, i)
with open(fname, 'wb') as f:
dataset.save_xml(f)
reduction['fname'] = fnames
return deepcopy(reduction)
def resolution_kernel(p_theta,
p_wavelength,
theta0,
wavelength0,
npnts=1001,
spectrum=None):
"""
Creates a full resolution kernel based on angular and wavelength components
Parameters
----------
p_theta: P_Theta
Angular component
p_wavelength: P_Wavelength
Wavelength components
theta0: array-like
Nominal angle of incidence, degrees
wavelength0: array-like
Nominal wavelength, Angstrom
npnts: float
number of points in the resolution kernel
spectrum: None or callable
Function, spectrum(wavelength) that specifies the intensity of
the neutron spectrum at a given wavelength
Returns
-------
kernel: np.ndarray
Full resolution kernel. Has shape `(N, 2, npnts)` where `N` is the
number of points in theta0/wavelength0.
kernel[:, 0, :] and kernel[:, 1, :] correspond to `Q` and `PDF(Q)` for
each of the data points in the first dimension.
"""
theta0_arr = np.asfarray(theta0).ravel()
wavelength0_arr = np.asfarray(wavelength0).ravel()
qpnts = max(theta0_arr.size, wavelength0_arr.size)
arr = [
np.array(a) for a in np.broadcast_arrays(theta0_arr, wavelength0_arr)
]
theta0_arr = arr[0]
wavelength0_arr = arr[1]
mean_q = general.q(theta0_arr, wavelength0_arr)
max_q = general.q(
theta0_arr + p_theta.width,
wavelength0_arr - p_wavelength.width(wavelength0_arr),
)
width = max_q - mean_q
kernel = np.zeros((qpnts, 2, npnts))
for i in range(qpnts):
Q = np.linspace(mean_q[i] - width[i], mean_q[i] + width[i], npnts)
kernel[i, 0, :] = Q
# angular component
pqt = pq_theta(p_theta, theta0_arr[i], wavelength0_arr[i], Q)
# burst time component
pqb = pq_wavelength(p_wavelength.burst, theta0_arr[i],
wavelength0_arr[i], Q, spectrum)
# crossing time component
pqc = pq_wavelength(
p_wavelength.crossing,
theta0_arr[i],
wavelength0_arr[i],
Q,
spectrum,
)
# rebinning component
pqda = pq_wavelength(p_wavelength.da, theta0_arr[i],
wavelength0_arr[i], Q, spectrum)
spacing = np.diff(Q)[0]
p = np.convolve(pqt, pqb, "same")
p = np.convolve(p, pqc, "same")
p = np.convolve(p, pqda, "same")
p *= spacing**3.0
kernel[i, 1, :] = p / integrate.simps(p, Q)
return kernel
def _reduce_single_angle(self, scale=1):
"""
Reduce a single angle.
"""
n_spectra = self.reflected_beam.n_spectra
n_tpixels = np.size(self.reflected_beam.m_topandtail, 1)
n_xpixels = np.size(self.reflected_beam.m_topandtail, 2)
# we'll need the wavelengths to calculate Q.
wavelengths = self.reflected_beam.m_lambda
m_twotheta = np.zeros((n_spectra, n_tpixels, n_xpixels))
detrot_difference = (self.reflected_beam.detector_z -
self.direct_beam.detector_z)
# difference in pixels between reflected position and direct beam
# at the two different detrots.
QZ_PIXEL_SPACING = self.reflected_beam.cat.qz_pixel_size[0]
dy = self.reflected_beam.detector_y
# convert that pixel difference to angle (in small angle approximation)
# higher `som` leads to lower m_beampos. i.e. higher two theta
# is at lower pixel values
beampos_2theta_diff = -(self.reflected_beam.m_beampos -
self.direct_beam.m_beampos)
beampos_2theta_diff *= QZ_PIXEL_SPACING / dy[0]
beampos_2theta_diff = np.degrees(beampos_2theta_diff)
total_2theta_deflection = detrot_difference + beampos_2theta_diff
# omega_nom.shape = (N, )
omega_nom = total_2theta_deflection / 2.0
omega_corrected = omega_nom[:, np.newaxis]
m_twotheta += np.arange(n_xpixels * 1.0)[np.newaxis, np.newaxis, :]
m_twotheta -= self.direct_beam.m_beampos[:, np.newaxis, np.newaxis]
# minus sign in following line because higher two theta is at lower
# pixel values
m_twotheta *= -QZ_PIXEL_SPACING / dy[:, np.newaxis, np.newaxis]
m_twotheta = np.degrees(m_twotheta)
m_twotheta += detrot_difference
# you may be reflecting upside down, reverse the sign.
upside_down = np.sign(omega_corrected[:, 0])
m_twotheta *= upside_down[:, np.newaxis, np.newaxis]
omega_corrected *= upside_down[:, np.newaxis]
"""
--Specular Reflectivity--
Use the (constant wavelength) spectra that have already been integrated
over 2theta (in processnexus) to calculate the specular reflectivity.
Beware: this is because m_topandtail has already been divided through
by monitor counts and error propagated (at the end of processnexus).
Thus, the 2theta pixels are correlated to some degree. If we use the 2D
plot to calculate reflectivity
(sum {Iref_{2theta, lambda}}/I_direct_{lambda}) then the error bars in
the reflectivity turn out much larger than they should be.
"""
ydata, ydata_sd = EP.EPdiv(
self.reflected_beam.m_spec,
self.reflected_beam.m_spec_sd,
self.direct_beam.m_spec,
self.direct_beam.m_spec_sd,
)
# calculate the 1D Qz values.
xdata = general.q(omega_corrected, wavelengths)
xdata_sd = (self.reflected_beam.m_lambda_fwhm /
self.reflected_beam.m_lambda)**2
xdata_sd += (self.reflected_beam.domega[:, np.newaxis] /
omega_corrected)**2
xdata_sd = np.sqrt(xdata_sd) * xdata
"""
---Offspecular reflectivity---
normalise the counts in the reflected beam by the direct beam
spectrum this gives a reflectivity. Also propagate the errors,
leaving the fractional variance (dr/r)^2.
--Note-- that adjacent y-pixels (same wavelength) are correlated in
this treatment, so you can't just sum over them.
i.e. (c_0 / d) + ... + c_n / d) != (c_0 + ... + c_n) / d
"""
m_ref, m_ref_sd = EP.EPdiv(
self.reflected_beam.m_topandtail,
self.reflected_beam.m_topandtail_sd,
self.direct_beam.m_spec[:, :, np.newaxis],
self.direct_beam.m_spec_sd[:, :, np.newaxis],
)
# you may have had divide by zero's.
m_ref = np.where(np.isinf(m_ref), 0, m_ref)
m_ref_sd = np.where(np.isinf(m_ref_sd), 0, m_ref_sd)
# calculate the Q values for the detector pixels. Each pixel has
# different 2theta and different wavelength, ASSUME that they have the
# same angle of incidence
qx, qy, qz = general.q2(
omega_corrected[:, :, np.newaxis],
m_twotheta,
0,
wavelengths[:, :, np.newaxis],
)
reduction = {}
reduction["x"] = self.x = xdata
reduction["x_err"] = self.x_err = xdata_sd
reduction["y"] = self.y = ydata / scale
reduction["y_err"] = self.y_err = ydata_sd / scale
reduction["omega"] = omega_corrected
reduction["m_twotheta"] = m_twotheta
reduction["m_ref"] = self.m_ref = m_ref
reduction["m_ref_err"] = self.m_ref_err = m_ref_sd
reduction["qz"] = self.m_qz = qz
reduction["qx"] = self.m_qx = qx
reduction["nspectra"] = self.n_spectra = n_spectra
reduction["start_time"] = self.reflected_beam.start_time
reduction[
"datafile_number"] = self.datafile_number = self.reflected_beam.datafile_number
fnames = []
datasets = []
datafilename = self.reflected_beam.datafilename
datafilename = os.path.basename(datafilename.split(".nx.hdf")[0])
for i in range(n_spectra):
data_tup = self.data(scanpoint=i)
datasets.append(ReflectDataset(data_tup))
if self.save:
for i, dataset in enumerate(datasets):
fname = "{0}_{1}.dat".format(datafilename, i)
fnames.append(fname)
with open(fname, "wb") as f:
dataset.save(f)
fname = "{0}_{1}.xml".format(datafilename, i)
with open(fname, "wb") as f:
dataset.save_xml(f, start_time=reduction["start_time"][i])
reduction["fname"] = fnames
return datasets, deepcopy(reduction)
def sample(self, samples, random_state=None):
"""
Sample the beam for reflected signal.
2400000 samples roughly corresponds to 1200 sec of *PLATYPUS* using
dlambda=3.3 and dtheta=3.3 at angle=0.65.
150000000 samples roughly corresponds to 3600 sec of *PLATYPUS* using
dlambda=3.3 and dtheta=3.3 at angle=3.0.
(The sample number <--> actual acquisition time correspondence has
not been checked fully)
Parameters
----------
samples: int
How many samples to run.
random_state: {int, `~np.random.RandomState`, `~np.random.Generator`}, optional
If `random_state` is not specified the
`~np.random.RandomState` singleton is used.
If `random_state` is an int, a new ``RandomState`` instance is used,
seeded with seed.
If `random_state` is already a ``RandomState`` or a ``Generator``
instance, then that object is used.
Specify `random_state` for repeatable minimizations.
"""
# grab a random number generator
rng = check_random_state(random_state)
# generate neutrons of various wavelengths
wavelengths = self.spectrum_dist.rvs(size=samples, random_state=rng)
# generate neutrons of different angular divergence
angles = self.angular_dist.rvs(samples, random_state=rng) + self.angle
# angular deviation due to gravity
# --> no correction for gravity affecting width of angular resolution
if self.gravity:
speeds = general.wavelength_velocity(wavelengths)
# trajectories through slits for different wavelengths
trajectories = pm.find_trajectory(self.L12 / 1000.0, 0, speeds)
# elevation at sample
elevations = pm.elevation(
trajectories, speeds, (self.L12 + self.L2S) / 1000.0
)
angles -= elevations
# calculate Q
q = general.q(angles, wavelengths)
# calculate reflectivities for a neutron of a given Q.
# the angular resolution smearing has already been done. The wavelength
# resolution smearing follows.
r = self.model(q, x_err=0.0)
# accept or reject neutrons based on the reflectivity of
# sample at a given Q.
criterion = rng.uniform(size=samples)
accepted = criterion < r
# implement wavelength smearing from choppers. Jitter the wavelengths
# by a uniform distribution whose full width is dlambda / 0.68.
if self.force_gaussian:
noise = rng.standard_normal(size=samples)
jittered_wavelengths = wavelengths * (
1 + self.dlambda / 2.3548 * noise
)
else:
noise = rng.uniform(-0.5, 0.5, size=samples)
jittered_wavelengths = wavelengths * (
1 + self.dlambda / 0.68 * noise
)
# update reflected beam counts. Rebin smearing
# is taken into account due to the finite size of the wavelength
# bins.
hist = np.histogram(
jittered_wavelengths[accepted], self.wavelength_bins
)
self.reflected_beam += hist[0]
self.bmon_reflect += float(samples)
# update resolution kernel. If we have more than 100000 in all
# bins skip
if (
len(self._res_kernel)
and np.min([len(v) for v in self._res_kernel.values()]) > 500000
):
return
bin_loc = np.digitize(jittered_wavelengths, self.wavelength_bins)
for i in range(1, len(self.wavelength_bins)):
# extract q values that fall in each wavelength bin
q_for_bin = np.copy(q[bin_loc == i])
q_samples_so_far = self._res_kernel.get(i - 1, np.array([]))
updated_samples = np.concatenate((q_samples_so_far, q_for_bin))
# no need to keep double precision for these sample arrays
self._res_kernel[i - 1] = updated_samples.astype(np.float32)
def __init__(
self,
model,
angle,
L12=2859,
footprint=60,
L2S=120,
dtheta=3.3,
lo_wavelength=2.8,
hi_wavelength=18,
dlambda=3.3,
rebin=2,
gravity=False,
force_gaussian=False,
force_uniform_wavelength=False,
):
self.model = model
self.bkg = model.bkg.value
self.angle = angle
# dlambda refers to the FWHM of the gaussian approximation to a uniform
# distribution. The full width of the uniform distribution is
# dlambda/0.68.
self.dlambda = dlambda / 100.0
# the rebin percentage refers to the full width of the bins. You have to
# multiply this value by 0.68 to get the equivalent contribution to the
# resolution function.
self.rebin = rebin / 100.0
self.wavelength_bins = calculate_wavelength_bins(
lo_wavelength, hi_wavelength, rebin
)
bin_centre = 0.5 * (
self.wavelength_bins[1:] + self.wavelength_bins[:-1]
)
# angular deviation due to gravity
# --> no correction for gravity affecting width of angular resolution
elevations = 0
if gravity:
speeds = general.wavelength_velocity(bin_centre)
# trajectories through slits for different wavelengths
trajectories = pm.find_trajectory(L12 / 1000.0, 0, speeds)
# elevation at sample
elevations = pm.elevation(
trajectories, speeds, (L12 + L2S) / 1000.0
)
# nominal Q values
self.q = general.q(angle - elevations, bin_centre)
# keep a tally of the direct and reflected beam
self.direct_beam = np.zeros((self.wavelength_bins.size - 1))
self.reflected_beam = np.zeros((self.wavelength_bins.size - 1))
# beam monitor counts for normalisation
self.bmon_direct = 0
self.bmon_reflect = 0
self.gravity = gravity
# wavelength generator
self.force_uniform_wavelength = force_uniform_wavelength
if force_uniform_wavelength:
self.spectrum_dist = uniform(
loc=lo_wavelength - 1, scale=hi_wavelength - lo_wavelength + 1
)
else:
a = PN("PLP0000711.nx.hdf")
q, i, di = a.process(
normalise=False,
normalise_bins=False,
rebin_percent=0.5,
lo_wavelength=max(0, lo_wavelength - 1),
hi_wavelength=hi_wavelength + 1,
)
q = q.squeeze()
i = i.squeeze()
self.spectrum_dist = SpectrumDist(q, i)
self.force_gaussian = force_gaussian
# angular resolution generator, based on a trapezoidal distribution
# The slit settings are the optimised set typically used in an
# experiment. dtheta/theta refers to the FWHM of a Gaussian
# approximation to a trapezoid.
# stores the q vectors contributing towards each datapoint
self._res_kernel = {}
self._min_samples = 0
self.dtheta = dtheta / 100.0
self.footprint = footprint
self.angle = angle
self.L2S = L2S
self.L12 = L12
s1, s2 = general.slit_optimiser(
footprint,
self.dtheta,
angle=angle,
L2S=L2S,
L12=L12,
verbose=False,
)
div, alpha, beta = general.div(s1, s2, L12=L12)
self.div, self.s1, self.s2 = s1, s2, div
if force_gaussian:
self.angular_dist = norm(scale=div / 2.3548)
else:
self.angular_dist = trapz(
c=(alpha - beta) / 2.0 / alpha,
d=(alpha + beta) / 2.0 / alpha,
loc=-alpha,
scale=2 * alpha,
)
def test_q(self):
q = general.q(1., 2.)
assert_almost_equal(q, 0.1096567037)
def test_q(self):
q = general.q(1., 2.)
assert_almost_equal(q, 0.1096567037)
