def generate_LAL_modes(approximant, q, chiA0, chiB0, dt, M, \
dist_mpc, f_low, f_ref, phi_ref, ellMax=None):
distance = dist_mpc * 1.0e6 * PC_SI
approxTag = lalsim.SimInspiralGetApproximantFromString(approximant)
# component masses of the binary
m1_kg = M * MSUN_SI * q / (1. + q)
m2_kg = M * MSUN_SI / (1. + q)
dictParams = lal.CreateDict()
if ellMax is not None:
ma = lalsim.SimInspiralCreateModeArray()
for ell in range(2, ellMax + 1):
lalsim.SimInspiralModeArrayActivateAllModesAtL(ma, ell)
lalsim.SimInspiralWaveformParamsInsertModeArray(dictParams, ma)
lmax = 5 # This in unused
hmodes = lalsim.SimInspiralChooseTDModes(phi_ref, dt, m1_kg, m2_kg, \
chiA0[0], chiA0[1], chiA0[2], chiB0[0], chiB0[1], chiB0[2], \
f_low, f_ref, distance, dictParams, lmax, approxTag)
t = np.arange(len(hmodes.mode.data.data)) * dt
mode_dict = {}
while hmodes is not None:
mode_dict['h_l%dm%d' % (hmodes.l, hmodes.m)] = hmodes.mode.data.data
hmodes = hmodes.next
return t, mode_dict
def get_SEOBNRv4PHM_modes(q, M, chi1, chi2, f_start, distance, deltaT):
"""Generate SEOBNRv4PHM modes"""
m1SI = lal.MSUN_SI * q * M / (1.0 + q)
m2SI = lal.MSUN_SI * M / (1.0 + q)
approx = ls.SEOBNRv4PHM
hlm = ls.SimInspiralChooseTDModes(
0.,
deltaT,
m1SI,
m2SI,
chi1[0],
chi1[1],
chi1[2],
chi2[0],
chi2[1],
chi2[2],
f_start,
f_start,
distance,
None,
5,
approx,
)
hI = {}
modes = [(2, 2), (2, 1), (3, 3), (4, 4), (5, 5)]
for lm in modes:
hI[lm] = ls.SphHarmTimeSeriesGetMode(hlm, lm[0], lm[1]).data.data
return hI
def load_lvcnr_data(filepath, mode, M, dt, inclination, phiRef, \
dist_mpc, f_low=0):
""" If f_low = 0, uses the entire NR data. The actual f_low will be
returned.
"""
NRh5File = h5py.File(filepath, 'r')
# set mode for NR data
params_NR = lal.CreateDict()
lalsim.SimInspiralWaveformParamsInsertNumRelData(params_NR, filepath)
if mode != 'all':
params_NR = set_single_mode(params_NR, mode[0], mode[1])
# Metadata parameters masses:
m1 = NRh5File.attrs['mass1']
m2 = NRh5File.attrs['mass2']
m1SI = m1 * M / (m1 + m2) * MSUN_SI
m2SI = m2 * M / (m1 + m2) * MSUN_SI
distance = dist_mpc * 1.0e6 * PC_SI
f_ref = f_low
spins = lalsim.SimInspiralNRWaveformGetSpinsFromHDF5File(f_ref, M, \
filepath)
s1x = spins[0]
s1y = spins[1]
s1z = spins[2]
s2x = spins[3]
s2y = spins[4]
s2z = spins[5]
# If f_low == 0, update it to the start frequency so that the surrogate
# gets the right start frequency
if f_low == 0:
f_low = NRh5File.attrs['f_lower_at_1MSUN'] / M
f_ref = f_low
f_low = f_ref
lmax = 5
# Generating the NR waveform
approx = lalsim.NR_hdf5
hmodes = lalsim.SimInspiralChooseTDModes(phiRef, dt, m1SI, m2SI, \
s1x, s1y, s1z, s2x, s2y, s2z, \
f_low, f_ref, distance, params_NR, lmax, approx)
t = np.arange(len(hmodes.mode.data.data)) * dt
mode_dict = {}
while hmodes is not None:
mode_dict['h_l%dm%d' % (hmodes.l, hmodes.m)] = hmodes.mode.data.data
hmodes = hmodes.next
return t, mode_dict, q, s1x, s1y, s1z, s2x, s2y, s2z, f_low, f_ref
'''
def get_SEOBNRv4PHM_modes(q, M, chi1, chi2, f_start, deltaT=1. / (4096 * 4.)):
#See the paper: https://arxiv.org/pdf/2004.09442.pdf
"""Generate SEOBNRv4PHM modes"""
prefactor = 4.7864188273360336e-20 # G/c^2*(M_sun/Mpc)
distance = 1. * 1e6 * lal.PC_SI # 1 Mpc in m
amp_prefactor = prefactor * M / 1. # G/c^2 (M / d_L)
nu = q / (1 + q)**2
m1SI = lal.MSUN_SI * q * M / (1.0 + q)
m2SI = lal.MSUN_SI * M / (1.0 + q)
#approx = lalsim.SEOBNRv4PHM
approx = lalsim.SimInspiralGetApproximantFromString('SEOBNRv4PHM')
hlm = lalsim.SimInspiralChooseTDModes(
0.,
deltaT,
m1SI,
m2SI,
chi1[0],
chi1[1],
chi1[2],
chi2[0],
chi2[1],
chi2[2],
f_start, #/**< starting GW frequency (Hz) */ #what is this f_start??
f_start, #/**< reference GW frequency (Hz) */
distance,
None,
5,
approx,
)
hI = {}
modes = [(2, 2), (2, 1), (2, -1), (3, 3), (4, 4), (5, 5)]
for lm in modes:
hI[lm] = lalsim.SphHarmTimeSeriesGetMode(
hlm, lm[0], lm[1]).data.data / amp_prefactor / nu
times = np.linspace(0, len(hI[(2, 2)]) * deltaT, len(hI[(2, 2)]))
h_22 = hI[(2, 2)]
ph = np.unwrap(np.angle(np.conj(h_22)))
omega_22 = (ph[2] - ph[0]) / (2 * deltaT)
print("FREQUENCY of THE WF: ", omega_22, omega_22 / (2 * np.pi))
t_max = times[np.argmax(np.abs(h_22))]
times = times - t_max
return times, hI
def get_SEOBNRv4PHM_modes(q, M, chi1, chi2, f_start, deltaT):
"""Generate SEOBNRv4PHM modes"""
m1SI = lal.MSUN_SI * q * M / (1.0 + q)
m2SI = lal.MSUN_SI * M / (1.0 + q)
approx = ls.SEOBNRv4PHM
prefactor = 4.7864188273360336e-20 # G/c^2*(M_sun/Mpc)
amp_prefactor = prefactor * M / 1. # G/c^2 (M / d_L)
nu = np.divide(q, np.square(1 + q))
distance = 1. * 1e6 * lal.PC_SI
hlm = ls.SimInspiralChooseTDModes(0., deltaT, m1SI, m2SI, chi1[0], chi1[1],
chi1[2], chi2[0], chi2[1], chi2[2],
f_start, f_start, distance, None, 5,
approx)
hI = {}
modes = [(2, 2), (2, 1), (3, 3), (4, 4), (5, 5)]
for lm in modes:
hI[lm] = ls.SphHarmTimeSeriesGetMode(
hlm, lm[0], lm[1]).data.data / (amp_prefactor * nu)
t = np.linspace(0, deltaT * len(hI[(2, 2)]), len(hI[(2, 2)]))
t = t - t[np.argmax(np.abs(hI[(2, 2)]))]
return t, hI
h_P_mlgw_T = twist_modes(g, h_22_NP, h_21_NP, alpha, beta,
gamma) #here mlgw performs only the twist
t_grid = t_grid - t_grid[np.argmax(h_22_NP)]
#P choose TD modes (in L0 frame)
#The change of frame is performed in XLALSimIMRPhenomTPHM_L0Modes (https://git.ligo.org/lscsoft/lalsuite/-/blob/master/lalsimulation/lib/LALSimIMRPhenomTPHM.c#L328)
hlm = lalsim.SimInspiralChooseTDModes(
0.,
t_step,
theta[0, 0] * lalsim.lal.MSUN_SI,
theta[0, 1] * lalsim.lal.MSUN_SI,
theta[0, 2],
theta[0, 3],
theta[0, 4],
theta[0, 5],
0.,
theta[0, 7],
f_min,
f_min,
1e6 * lalsim.lal.PC_SI,
lal.CreateDict(),
5, #lmax
lalsim.IMRPhenomTPHM)
prefactor = 4.7864188273360336e-20
m1, m2 = theta[0, 0], theta[0, 1]
nu = np.divide(m1 / m2, np.square(1 + m1 / m2))
amp_prefactor = prefactor * (m1 + m2) / 1. * nu
h_22_P_lal = lalsim.SphHarmTimeSeriesGetMode(hlm, 2,
2).data.data / amp_prefactor
h_21_P_lal = lalsim.SphHarmTimeSeriesGetMode(hlm, 2,
def create_dataset_TD(N_data,
N_grid,
modes,
basefilename,
t_coal=0.5,
q_range=(1., 5.),
m2_range=None,
s1_range=(-0.8, 0.8),
s2_range=(-0.8, 0.8),
t_step=1e-5,
alpha=0.35,
approximant="SEOBNRv2_opt",
path_TEOBResumS=None):
"""
create_dataset_TD
=================
Create datasets for training a ML model to fit GW modes in time domain. Each dataset is saved in a different file (basefilename.mode).
The dataset consists in 3 parameters theta=(q, spin1z, spin2z) associated to the modes computed in time domain for a grid of N_grid points in the range given by the user.
More specifically, data are stored in 3 vectors:
theta_vector vector holding source parameters q, spin1, spin2
amp_vector vector holding amplitudes for each source evaluated at some discrete grid points
ph_vector vector holding phase for each source evaluated at some discrete grid points
This routine add N_data data to filename if one is specified (if file is not empty it must contain data with the same N_grid); otherwise the datasets are returned as np vectors.
Values of q and m2 as well as spins are drawn randomly in the range given by the user: it holds m1 = q *m2 M_sun.
The waveforms are computed with a time step t_step; starting from a time in reduced grid tau min (s/M_Sun). Waves are given in a rescaled time grid (i.e. t/m_tot) with N_grid points: t=0 occurs when the 22 mode has a peak. A higher density of grid points is placed close to the merger.
Dataset is generated either with an implementation of TEOBResumS (a path to a local installation of TEOBResumS should be provided) either with SEOBNRv4HM (lalsuite installation required). It can be given an TD lal approximant with no HM; in this case, only the 22 mode can be generated.
Datasets can be loaded with load_dataset.
Input:
N_data size of dataset
N_grid number of grid points to evaluate
modes [] list of modes (each a (l,m) tuple) to generate and fill the dataset with
basefilename basename of the file to save dataset in (each dataset is saved in basefilename.lm)
t_coal time to coalescence to start computation from (measured in reduced grid)
q_range tuple with range for random q values. if single value, q is kept fixed at that value
m2_range tuple with range for random m2 values. if single value, m2 is kept fixed at that value. If None, m2 will be chosen s.t. m_tot = m1+m2 = 20. M_sun
spin_mag_max_1 tuple with range for random spin #1 values. if single value, s1 is kept fixed at that value
spin_mag_max_2 tuple with range for random spin #1 values. if single value, s2 is kept fixed at that value
t_step time step to generate the wave with
approximant string for the approximant model to be used (in lal convention; to be used only if lal is to be used)
alpha distorsion factor for time grid. (In range (0,1], when it's close to 0, more grid points are around merger)
approximant lal approximant to be used for generating the modes, or "TEOBResumS" (in the latter case a local installation must be provided by the argument path_TEOBResumS)
path_TEOBResumS path to a local installation of TEOBResumS with routine 'EOBRun_module' (has effect only if approximant is "TEOBResumS")
"""
#TODO: create a function get modes, wrapper to ChooseTDModes: you can call it from here to get the modes
if isinstance(modes, tuple):
modes = [modes]
if not isinstance(modes, list):
raise TypeError(
"Wrong kind of mode {} given. Expected a list [(l,m)]".format(
modes))
if approximant == "TEOBResumS":
#see https://bitbucket.org/eob_ihes/teobresums/src/development/ for the implementation of TEOBResumS
if not isinstance(path_TEOBResumS, str):
raise ValueError("Missing path to TEOBResumS: unable to continue")
try:
import sys
sys.path.append(
path_TEOBResumS) #path to local installation of TEOBResumS
import EOBRun_module
except:
raise RuntimeError(
"No valid imput source for module 'EOBRun_module' for TEOBResumS. Unable to continue."
)
else:
try:
import lal
import lalsimulation as lalsim
except:
raise RuntimeError(
"Impossible to load lal and lalsimulation: try pip install lalsuite"
)
LALpars = lal.CreateDict()
approx = lalsim.SimInspiralGetApproximantFromString(approximant)
prefactor = 4.7864188273360336e-20 # G/c^2*(M_sun/Mpc)
#checking that all is good with modes
if approximant == "SEOBNRv4HM":
for mode in modes:
if mode not in [(2, 2), (2, 1), (3, 3), (4, 4), (5, 5)]:
raise ValueError(
"Currently SEOBNRv4HM approximants do not implement the chosen HM"
)
elif approximant != "IMRPhenomTPHM":
if modes != [(2, 2)]:
raise ValueError(
"The chosen lal approximant does not implement HMs")
#checking if N_grid is fine
if not isinstance(N_grid, int):
raise TypeError("N_grid is " + str(type(N_grid)) +
"! Expected to be a int.")
if isinstance(m2_range, tuple):
D_theta = 4 #m2 must be included as a feature
else:
D_theta = 3
######setting the time grid
time_grid_list = []
t_end_list = []
if approximant == "TEOBResumS":
modes_to_k = lambda modes: [
int(x[0] * (x[0] - 1) / 2 + x[1] - 2) for x in modes
] # [(l,m)] -> [k]
k_modes = modes_to_k(modes)
#setting a list of time grids
for mode in modes:
#ugly setting of t_end in TEOBResumS: required to kill bad features after merger
if mode == (2, 2):
t_end = 5.2e-4 #estimated maximum time for ringdown: WF will be killed after that time
elif mode == (2,
1) or mode == (3,
3): #special treatment for 21 and 33
t_end = 1e-6
else:
t_end = 3e-5 #for other modes
t_end_list.append(t_end)
else:
for mode in modes:
t_end_list.append(5.2e-4)
print("Generating modes: " + str(modes))
#creating time_grid
for i, mode in enumerate(modes):
time_grid = np.linspace(-np.power(np.abs(t_coal), alpha),
np.power(t_end_list[i], alpha), N_grid)
time_grid = np.multiply(np.sign(time_grid),
np.power(np.abs(time_grid), 1. / alpha))
#adding 0 to time grid
index_0 = np.argmin(np.abs(time_grid))
time_grid[index_0] = 0. #0 is alway set in the grid
time_grid_list.append(time_grid)
#setting t_coal_freq for generating a waves
if np.abs(t_coal) < 0.05:
t_coal_freq = 0.05
else:
t_coal_freq = np.abs(t_coal)
#####create a list of buffer to save the WFs
buff_list = []
for i, mode in enumerate(modes):
filename = basefilename + '.' + str(mode[0]) + str(mode[1])
if not os.path.isfile(
filename
): #file doesn't exist: must be created with proper header
filebuff = open(filename, 'w')
print("New file ", filename, " created")
time_header = np.concatenate((np.zeros(
(3, )), time_grid_list[i], time_grid_list[i]))[None, :]
np.savetxt(filebuff,
time_header,
header="#Mode:" + str(mode[0]) + str(mode[1]) +
"\n# row: theta " + str(D_theta) + " | amp (None," +
str(N_grid) + ")| ph (None," + str(N_grid) +
")\n# N_grid = " + str(N_grid) + " | t_coal =" +
str(t_coal) + " | t_step =" + str(t_step) +
" | q_range = " + str(q_range) + " | m2_range = " +
str(m2_range) + " | s1_range = " + str(s1_range) +
" | s2_range = " + str(s2_range),
newline='\n')
else:
filebuff = open(filename, 'a')
buff_list.append(filebuff)
#####creating WFs
for n_WF in range(N_data):
#setting value for data
if isinstance(m2_range, tuple):
m2 = np.random.uniform(m2_range[0], m2_range[1])
elif m2_range is not None:
m2 = float(m2_range)
if isinstance(q_range, tuple):
q = np.random.uniform(q_range[0], q_range[1])
else:
q = float(q_range)
if isinstance(s1_range, tuple):
spin1z = np.random.uniform(s1_range[0], s1_range[1])
else:
spin1z = float(s1_range)
if isinstance(s2_range, tuple):
spin2z = np.random.uniform(s2_range[0], s2_range[1])
else:
spin2z = float(s2_range)
if m2_range is None:
m2 = 20. / (1 + q)
m1 = q * m2
else:
m1 = q * m2
nu = np.divide(q, np.square(1 + q)) #symmetric mass ratio
#computing f_min
f_min = .9 * ((151 * (t_coal_freq)**(-3. / 8.) *
(((1 + q)**2) / q)**(3. / 8.)) / (m1 + m2))
#in () there is the right scaling formula for frequency in order to get always the right reduced time
#this should be multiplied by a prefactor (~1) for dealing with some small variation due to spins
if isinstance(m2_range, tuple):
temp_theta = [m1, m2, spin1z, spin2z]
else:
temp_theta = [m1 / m2, spin1z, spin2z]
#getting the wave
#output of the if:
#amp_list, ph_list (same order as in modes)
#time_full, argpeak
amp_list, ph_list = [None for i in range(len(modes))
], [None for i in range(len(modes))]
if approximant == "TEOBResumS": #using TEOBResumS
pars = {
'M': m1 + m2,
'q': m1 / m2,
'Lambda1': 0.,
'Lambda2': 0.,
'chi1': spin1z,
'chi2': spin2z,
'domain': 0, # TD
'arg_out': 1, # Output hlm/hflm. Default = 0
'use_mode_lm': list(
set(k_modes + [1])
), # List of modes to use/output through EOBRunPy (added 22 mode in case there isn't)
#'srate_interp' : 1./t_step, # srate at which to interpolate. Default = 4096.
'use_geometric_units':
0, # Output quantities in geometric units. Default = 1
'initial_frequency':
f_min, # in Hz if use_geometric_units = 0, else in geometric units
'interp_uniform_grid':
0, # Interpolate mode by mode on a uniform grid. Default = 0 (no interpolation)
'distance': 1.,
'inclination': 0.,
}
time_full, h_p, h_c, hlm = EOBRun_module.EOBRunPy(pars)
for i, k_mode in enumerate(k_modes):
temp_amp = hlm[str(k_mode)][
0] * nu #TEOBResumS has weird conventions on the modes
temp_ph = hlm[str(k_mode)][1]
amp_list[i] = temp_amp
ph_list[i] = temp_ph
argpeak = locate_peak(hlm['1'][0] *
nu) #aligned at the peak of the 22
elif approximant == "SEOBNRv4HM:old": #using SEOBNRv4HM
nqcCoeffsInput = lal.CreateREAL8Vector(10)
sp, dyn, dynHi = lalsim.SimIMRSpinAlignedEOBModes(
t_step, m1 * lal.MSUN_SI, m2 * lal.MSUN_SI, f_min,
1e6 * lal.PC_SI, spin1z, spin2z, 41, 0., 0., 0., 0., 0., 0.,
0., 0., 1., 1., nqcCoeffsInput, 0)
amp_prefactor = prefactor * (m1 + m2) / 1. # G/c^2 (M / d_L)
while sp is not None:
lm = (sp.l, sp.m)
if lm not in modes: #skipping a non-necessary mode
continue
else: #computing index and saving the mode
i = modes.index(lm)
hlm = sp.mode.data.data #complex mode vector
temp_amp = np.abs(
hlm
) / amp_prefactor / nu #scaling with the convention of SEOB
temp_ph = np.unwrap(np.angle(hlm))
amp_list[i] = temp_amp
ph_list[i] = temp_ph
if (sp.l, sp.m) == (2, 2): #get grid
amp_22 = np.abs(
sp.mode.data.data
) #amp of 22 mode (for time grid alignment)
time_full = np.linspace(
0.0, sp.mode.data.length * t_step, sp.mode.data.length
) #time grid at which wave is computed
argpeak = locate_peak(
amp_22) #aligned at the peak of the 22
sp = sp.next
elif approximant == "IMRPhenomTPHM" or approximant == "SEOBNRv4HM":
#https://lscsoft.docs.ligo.org/lalsuite/lalsimulation/test___s_e_o_b_n_rv4_p_h_m__vs__4_h_m__ringdown_8py_source.html
#approx = lalsim.SEOBNRv4PHM #DEBUG
hlm = lalsim.SimInspiralChooseTDModes(
0.,
t_step,
m1 * lalsim.lal.MSUN_SI,
m2 * lalsim.lal.MSUN_SI,
0.,
0.,
spin1z,
0.,
0.,
spin2z,
f_min,
f_min,
1e6 * lalsim.lal.PC_SI,
LALpars,
5, #lmax
lalsim.IMRPhenomTPHM)
amp_prefactor = prefactor * (m1 + m2) / 1.
for i, lm in enumerate(modes):
temp_hlm = lalsim.SphHarmTimeSeriesGetMode(hlm, lm[0],
lm[1]).data.data
temp_amp = np.abs(
temp_hlm
) / amp_prefactor / nu #check that this conventions are for every lal part
temp_ph = np.unwrap(np.angle(temp_hlm))
amp_list[i] = temp_amp
ph_list[i] = temp_ph
if (lm[0], lm[1]) == (2, 2): #get grid
argpeak = locate_peak(
temp_amp) #aligned at the peak of the 22
time_full = np.linspace(
0.0,
len(temp_amp) * t_step,
len(temp_amp)) #time grid at which wave is computed
else: #another lal approximant (only 22 mode)
hp, hc = lalsim.SimInspiralChooseTDWaveform( #where is its definition and documentation????
m1 * lalsim.lal.MSUN_SI, #m1
m2 * lalsim.lal.MSUN_SI, #m2
0.,
0.,
spin1z, #spin vector 1
0.,
0.,
spin2z, #spin vector 2
1e6 * lalsim.lal.PC_SI, #distance to source
0., #inclination
0., #phi
0., #longAscNodes
0., #eccentricity
0., #meanPerAno
t_step, # time incremental step
f_min, # lowest value of freq
f_min, #some reference value of freq (??)
LALpars, #some lal dictionary
approx #approx method for the model
)
h_p, h_c = hp.data.data, hc.data.data
time_full = np.linspace(
0.0, hp.data.length * t_step,
hp.data.length) #time grid at which wave is computed
amp_prefactor = prefactor * (m1 + m2) / 1. # G/c^2 (M / d_L)
temp_amp = np.sqrt(np.square(h_p) +
np.square(h_c)) / amp_prefactor / (
4 * np.sqrt(5 / (64 * np.pi)))
temp_ph = np.unwrap(np.arctan2(h_c, h_p))
amp_list = [temp_amp]
ph_list = [temp_ph]
argpeak = locate_peak(temp_amp) #aligned at the peak of the 22
#setting time grid
t_peak = time_full[argpeak]
time_full = (time_full - t_peak) / (m1 + m2
) #grid is scaled to standard grid
#computing waves to the chosen std grid and saving to file
for i in range(len(amp_list)):
temp_amp, temp_ph = amp_list[i], ph_list[i]
temp_amp = np.interp(time_grid_list[i], time_full, temp_amp)
temp_ph = np.interp(time_grid_list[i], time_full, temp_ph)
temp_ph = temp_ph - temp_ph[
0] #all phases are shifted by a constant to make sure every wave has 0 phase at beginning of grid
if False:
#TODO: remove this shit!!
plt.figure()
plt.plot(time_grid_list[i], temp_amp_, 'o', ms=2)
plt.plot(time_full, temp_amp)
plt.xlim([-0.00015, 0.00015])
plt.show()
to_save = np.concatenate(
(temp_theta, temp_amp, temp_ph))[None, :] #(1,D)
np.savetxt(buff_list[i], to_save)
#user communication
if n_WF % 50 == 0 and n_WF != 0:
#if n_WF%1 == 0 and n_WF != 0: #debug
print("Generated WF ", n_WF)
if filename is None:
return theta_vector, amp_dataset.real, ph_dataset.real, time_grid
else:
filebuff.close()
return None