Skip to content

GitLab

Commit e63aeb09 authored by Marcus Haberland's avatar Marcus Haberland
Browse files

changed the integration method to MC integration 

We exchanged scipy.integrate.quad with vegas integration
parent 09b222c0
Hide whitespace changes
Inline Side-by-side
binary.py
View file @ e63aeb09
......@@ -10,6 +10,7 @@ import matplotlib.pyplot as plt
from scipy import integrate, optimize
import LISA
import pickle
import vegas
pi = np.pi
AU = 1495978707e2 # m
......@@ -28,7 +29,6 @@ labels=['K','P','phi_0', 'theta_S_Ec', 'phi_S_Ec', 'theta_L', 'phi_L', 'ln(A)',
keys = enumerate(labels)
ArcTan = lambda x,y: np.arctan2(y,x)
file = 'dict_binaries.txt'
def isclose(a, b, rel_tol=1e-04, abs_tol=0.0):
return abs(a-b) <= max(rel_tol * max(abs(a), abs(b)), abs_tol)
......@@ -129,7 +129,7 @@ class binary:
self.gw_amplitude = self.a0*np.array([(1.+self.cos_inclination**2)/2, -self.cos_inclination]) # [A_plus, A_cross]
def sep(self):
return (G*(self.m1+self.m2)/(pi*freq)**2)**(1/6) #m
return (G*(self.m1+self.m2)/(pi*self.f_GW)**2)**(1/6) #m
def S_n(self):
return LISA.LISA().Sn(self.f_GW)
......@@ -263,16 +263,27 @@ class binary:
print('Error in d_strain')
raise ValueError
def h_ixh_j(self,i,j,diff=1e-6):
def h_ixh_j(self,i,j,diff=1e-6,num=10**4,epsrel=1e-2):
print('Integrating dh/d{} * dh/d{} ...\n'.format(labels[i],labels[j]))
if i == j:
funcA = self.h_i(i,diff)
return integrate.quad(lambda t: (funcA(t)*1e20)**2,0,self.T_obs,limit=int(self.T_obs*self.f_GW*20),epsrel=self.epsrel,epsabs=0)[0]*1e-40
integ = vegas.Integrator([[0,self.T_obs]])
result = integ(lambda t: (funcA(t))**2,nitn=10,neval=num,adapt=False)
it = 0
while(result[0].sdev/result[0].mean > epsrel and it < 1): # try again, but harder
it += 1
print('Trying again, because Q = {:.3f} and relative error = {:.3f}'.format(result.Q, result[0].sdev/result[0].mean))
result = integ(lambda t: (funcA(t))**2,nitn=10,neval=num*10**it,adapt=False)
print('Integral = {:.3e} +- {:.3e}, rel. uncertainty = {:.3f}, Q = {:.3f} \n'.format(result[0].mean, result[0].sdev,result[0].sdev/result[0].mean, result.Q))
return result[0].mean
if i != j:
funcA = self.h_i(i,diff)
funcB = self.h_i(j,diff)
return integrate.quad(lambda t: funcA(t)*1e20*funcB(t)*1e20,0,self.T_obs,limit=int(self.T_obs*self.f_GW*20),epsrel=self.epsrel,epsabs=np.sqrt(self.Fishers[self.key-1,i,i]*1e40*self.Fishers[self.key-1,j,j]*1e40)*self.epsabs)[0]*1e-40
integ = vegas.Integrator([[0,self.T_obs]])
result = integ(lambda t: funcA(t)*funcB(t),nitn=6,neval=num*10,adapt=False,rtol=epsrel,atol=np.sqrt(self.Fishers[self.key-1,i,i] * self.Fishers[self.key-1,j,j])*self.epsrel/10)
print(result.summary(),' \n Integral = {:.3e} +- {:.3e}, rel. uncertainty = {:.3f}, Q = {:.3f} \n'.format(result[0].mean, result[0].sdev, result[0].sdev/abs(result[0].mean), result.Q))
return result[0].mean
def h_i(self,i,diff=1e-6):
B = None
if i == 3:
......@@ -378,16 +389,25 @@ class binary:
self.Fisher *= 2./self.S_n() #2*mat/S_n
return self.Fisher
def rel_uncertainty(self,diff=1e-6,epsrel=1.49e-2,epsabs=1.49e-2,both=False):
def rel_uncertainty(self,diff=1e-6,both=False):
in_dict, js = self.in_dict_bin()
if in_dict:
return js['Tamanini_plot']
try:
unc = np.sqrt(np.abs(np.diag(np.linalg.inv(self.Fisher))[:3])) #np.linalg.eigvals(self.Fisher_mat()) for diagonalized
try:
unc = np.sqrt(np.diag(np.linalg.inv(self.Fisher)[:3]))
except:
print('Error in Uncertainty calculation! Error matrix isn´t positive on diagonal. Please retry with higher accuracy.')
unc = np.sqrt(np.abs(np.diag(np.linalg.inv(self.Fisher))[:3]))
self.add_json()
return unc/np.array([self.K,self.P,1.])
except:
unc = np.sqrt(np.abs(np.diag(np.linalg.inv(self.Fisher_mat(diff)))[:3])) #np.linalg.eigvals(self.Fisher_mat()) for diagonalized
try:
unc = np.sqrt(np.diag(np.linalg.inv(self.Fisher_mat(diff,both))[:3]))
except:
print('Error in Uncertainty calculation! Error matrix isn´t positive on diagonal. Please retry with higher accuracy.')
unc = np.sqrt(np.abs(np.diag(np.linalg.inv(self.Fisher_mat(diff,both)))[:3]))
unc = np.sqrt(np.abs(np.diag(np.linalg.inv(self.Fisher_mat(diff,both)))[:3])) #np.linalg.eigvals(self.Fisher_mat()) for diagonalized
self.add_json()
return unc/np.array([self.K,self.P,1.])
......@@ -398,43 +418,45 @@ class binary:
self.key = key
res += integrate.quad(lambda t: self.strain(t)**2,0,self.T_obs,limit=int(self.T_obs*self.f_GW*20))[0]
else:
res += 1e-20*integrate.quad(lambda t: (self.strain(t)*1e10)**2,0,self.T_obs,limit=int(self.T_obs*self.f_GW*20),epsrel=self.epsrel,epsabs=0)[0]
res += integrate.quad(lambda t: (self.strain(t))**2,0,self.T_obs,limit=int(self.T_obs*self.f_GW*20),epsrel=self.epsrel,epsabs=0)[0]
res *= np.sqrt(2)
return res
def signal_to_noise(self):
def signal_to_noise(self,both=False):
if not hasattr(self, 'snr'):
self.snr = np.sqrt(2*self.hxh()/self.S_n())
self.snr = np.sqrt(2*self.hxh(both)/self.S_n())
return self.snr
def json(self):
if not hasattr(self, 'Fisher'):
self.Fisher_mat()
err = np.linalg.inv(self.Fisher)
labels = ['K','P','phi_0', 'theta_S_Ec', 'phi_S_Ec', 'theta_L', 'phi_L', 'ln(A)', 'f_1', 'f_0','M_P','M_c','dist']
vals = [self.K,self.P/yr,self.phi_0,self.theta_S_Ec,self.phi_S_Ec,self.theta_L,self.phi_L,np.log(self.a0),self.f_1,self.f_GW,self.mP/M_J,self.chirp_mass,self.dist/pc]
if self.mode == 'm':
vals2 = [self.K,self.P,1.,1.,1.,1.,1.,np.log(self.a0),self.f_1]
if self.mode == 's':
vals2 = [self.K,self.P,1.]
if self.mode == 'l':
vals2 = [self.K,self.P,1.,1.,1.,1.,1.,np.log(self.a0),self.f_1,self.f_0]
rel_unc = err / np.outer(vals2,vals2)
self.js = {'Fisher' : self.Fisher,'Error' : err,
'pars' : {label : vals[i] for i,label in enumerate(labels)},
'correlation' : rel_unc,
'rel_unc' : np.sqrt(np.diag(rel_unc)),
'exo_rel_unc' : np.sqrt(np.diag(rel_unc))[:3],
'SNR' : self.signal_to_noise(),
'binary' : self
}
self.js['Tamanini_plot'] = self.js['SNR']*self.js['pars']['M_P']*self.js['exo_rel_unc']
if not hasattr(self, 'js'):
err = np.linalg.inv(self.Fisher)
labels = ['K','P','phi_0', 'theta_S_Ec', 'phi_S_Ec', 'theta_L', 'phi_L', 'ln(A)', 'f_1', 'f_0','M_P','M_c','dist']
vals = [self.K,self.P/yr,self.phi_0,self.theta_S_Ec,self.phi_S_Ec,self.theta_L,self.phi_L,np.log(self.a0),self.f_1,self.f_GW,self.mP/M_J,self.chirp_mass,self.dist/pc]
if self.mode == 'm':
vals2 = [self.K,self.P,1.,1.,1.,1.,1.,np.log(self.a0),self.f_1]
if self.mode == 's':
vals2 = [self.K,self.P,1.]
if self.mode == 'l':
vals2 = [self.K,self.P,1.,1.,1.,1.,1.,np.log(self.a0),self.f_1,self.f_0]
rel_unc = err / np.outer(vals2,vals2)
self.js = {'Fisher' : self.Fisher,'Error' : err,
'pars' : {label : vals[i] for i,label in enumerate(labels)},
'correlation' : rel_unc,
'rel_unc' : np.sqrt(np.diag(rel_unc)),
'exo_rel_unc' : np.sqrt(np.diag(rel_unc))[:3],
'SNR' : self.signal_to_noise(),
'binary' : self
}
self.js['Tamanini_plot'] = self.js['SNR']*self.js['pars']['M_P']*self.js['exo_rel_unc']
return self.js
def add_json(self):
in_dict, js = self.in_dict_bin()
if not in_dict:
self.json()
infile = open(file,'rb')
dict_binaries = pickle.load(infile)
infile.close()
......@@ -445,6 +467,7 @@ class binary:
outfile.close()
return self.js
else:
self.js = js
return js
def in_dict_bin(self):
......
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment