dev #295
@ -156,3 +156,20 @@ double energyDistCorrelation(double x, void *user_data) {
|
|||||||
return normalDist(x, e_m, e_b) * exp(-t * r);
|
return normalDist(x, e_m, e_b) * exp(-t * r);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// Generalised Gamma Function
|
||||||
|
double genGammaAlphaDist(double x, void *user_data) {
|
||||||
|
double *c = (double *)user_data;
|
||||||
|
|
||||||
|
double omega = c[0];
|
||||||
|
double tau0 = c[1];
|
||||||
|
double alpha = c[2];
|
||||||
|
double beta = c[3];
|
||||||
|
|
||||||
|
double b_to_a = beta / alpha;
|
||||||
|
double tau_to_tau0 = tau / tau0;
|
||||||
|
|
||||||
|
double norm = exp(-lgamma(b_to_a) + b_to_a * log(b_to_a)) * alpha;
|
||||||
|
|
||||||
|
return norm * exp(-b_to_a * pow(tau_to_tau0, alpha)) * pow(tau_to_tau0, beta);
|
||||||
|
}
|
||||||
|
|
||||||
|
@ -36,7 +36,10 @@ class AbstractGG(Distribution, ABC):
|
|||||||
taus, ln_tau = AbstractGG._prepare_integration(tau0)
|
taus, ln_tau = AbstractGG._prepare_integration(tau0)
|
||||||
g_tau = cls.distribution(taus, tau0, *args)
|
g_tau = cls.distribution(taus, tau0, *args)
|
||||||
|
|
||||||
ret_val = np.array([simpson(g_tau / (1 - 1j*w_i*taus), ln_tau) for w_i in w]).squeeze()
|
ret_val = np.zeros_like(omega, dtype=np.complex128)
|
||||||
|
|
||||||
|
ret_val.real = np.array([simpson(g_tau * taus / (1 + (w_i*taus)**2), ln_tau) for w_i in w]).squeeze()
|
||||||
|
ret_val.imag = np.array([simpson(g_tau * w_i * taus / (1 + (w_i*taus)**2), ln_tau) for w_i in w]).squeeze()
|
||||||
|
|
||||||
return ret_val
|
return ret_val
|
||||||
|
|
||||||
@ -50,7 +53,7 @@ class AbstractGG(Distribution, ABC):
|
|||||||
ret_val = np.zeros((w.size, _t.size))
|
ret_val = np.zeros((w.size, _t.size))
|
||||||
|
|
||||||
for i, tau_i in enumerate(_t):
|
for i, tau_i in enumerate(_t):
|
||||||
taus, ln_tau = AbstractGG._prepare_integration(tau_i)
|
taus, ln_tau = AbstractGG._prepare_integration(tau_i, limits=limits, num_steps=num_steps)
|
||||||
g_tau = cls.distribution(taus, tau_i, *args)
|
g_tau = cls.distribution(taus, tau_i, *args)
|
||||||
|
|
||||||
ret_val[:, i] = np.array([simpson(g_tau * taus / (1 + (w_i*taus)**2), ln_tau) for w_i in w]).squeeze()
|
ret_val[:, i] = np.array([simpson(g_tau * taus / (1 + (w_i*taus)**2), ln_tau) for w_i in w]).squeeze()
|
||||||
|
Loading…
Reference in New Issue
Block a user