def get_matrix_interp1d3_v2(x, xi):
N = len(xi) - 1
zeros = tf.zeros_like(x)
ones = tf.ones_like(x)
# @pysnooper.snoop()
def poly_i(i):
tmp = zeros
x_i = (xi[i] + xi[i - 1]) / 2
for j in range(i - 1, i + 3):
if j < 0 or j > N - 1:
continue
r = ones
for k in range(j - 1, j + 3):
if k == i or k < 1 or k > N:
continue
x_k = (xi[k] + xi[k - 1]) / 2
r = r * (x - x_k) / (x_i - x_k)
r = tf.where(
(x >= (xi[j] + xi[j - 1]) / 2) & (x < (xi[j] + xi[j + 1]) / 2),
r,
zeros,
)
tmp = tmp + r
return tmp
h = tf.stack([poly_i(i) for i in range(1, N)], axis=-1)
b = tf.zeros_like(x)
return h, b
def get_amp(self, data, _data_c=None, **kwargs):
mass = data["m"]
zeros = tf.zeros_like(mass)
a = tf.abs(self.a())
b = self.b()
r = -tf.complex(a, b) * tf.complex(mass * mass, zeros)
return tf.exp(r)
def interp(self, m):
# q = data_extra[self.outs[0]]["|q|"]
# a = self.a()
p = self.point_value()
zeros = tf.zeros_like(m)
ones = tf.ones_like(m)
def poly_i(i, xi):
tmp = zeros
for j in range(i - 1, i + 1):
if j < 0 or j > self.interp_N - 1:
continue
r = ones
for k in range(j, j + 2):
if k == i:
continue
r = r * (m - xi[k]) / (xi[i] - xi[k])
r = tf.where((m >= xi[j]) & (m < xi[j + 1]), r, zeros)
tmp = tmp + r
return tmp
h = tf.stack(
[poly_i(i, self.points) for i in range(1, self.interp_N - 1)],
axis=-1,
)
h = tf.stop_gradient(h)
p_r = tf.math.real(p)
p_i = tf.math.imag(p)
ret_r = tf.reduce_sum(h * p_r, axis=-1)
ret_i = tf.reduce_sum(h * p_i, axis=-1)
return tf.complex(ret_r, ret_i)
def cal_monentum(m, ma, mb):
mabp = ma + mb
mabm = ma - mb
s = m * m
p2 = (s - mabp * mabp) * (s - mabm * mabm) / 4 / s
zeros = tf.zeros_like(s)
p_p = tf.complex(tf.sqrt(tf.abs(p2)), zeros)
p_m = tf.complex(zeros, tf.sqrt(tf.abs(p2)))
return tf.where(p2 > 0, p_p, p_m)
def interp(self, m):
zeros = tf.zeros_like(m)
p = self.point_value()
xs = []
def poly_i(i):
x = 1.0
for j in range(self.interp_N):
if i == j:
continue
x = (x * (m - self.points[j]) /
(self.points[i] - self.points[j]))
return x
xs = tf.stack([poly_i(i) for i in range(self.interp_N)], axis=-1)
zeros = tf.zeros_like(xs)
xs = tf.complex(xs, zeros)
ret = tf.reduce_sum(xs[:, 1:-1] * p, axis=-1)
return ret
def interp(self, m):
zeros = tf.zeros_like(m)
p = self.point_value()
p_r = tf.math.real(p)
p_i = tf.math.imag(p)
xi_m = self.h_matrix
x_m = spline_x_matrix(m, self.points)
x_m = tf.expand_dims(x_m, axis=-1)
m_xi = tf.reduce_sum(xi_m * x_m, axis=[-3, -2])
m_xi = tf.stop_gradient(m_xi)
ret_r = tf.reduce_sum(tf.cast(m_xi, p_r.dtype) * p_r, axis=-1)
ret_i = tf.reduce_sum(tf.cast(m_xi, p_i.dtype) * p_i, axis=-1)
return tf.complex(ret_r, ret_i)
def interp(self, m):
p = self.point_value()
ones = tf.ones_like(m)
zeros = tf.zeros_like(m)
p_r = tf.math.real(p)
p_i = tf.math.imag(p)
h, b = get_matrix_interp1d3_v2(m, self.points)
h = tf.stop_gradient(h)
f = lambda x: tf.reshape(
tf.matmul(tf.cast(h, x.dtype), tf.reshape(x, (-1, 1))), b.shape
) + tf.cast(b, x.dtype)
ret_r = f(p_r)
ret_i = f(p_i)
return tf.complex(ret_r, ret_i)
def get_matrix_interp1d3(x, xi):
N = len(xi) - 1
zeros = tf.zeros_like(x)
ones = tf.ones_like(x)
# @pysnooper.snoop()
def poly_i(i):
tmp = zeros
for j in range(i - 1, i + 3):
if j < 0 or j > N - 1:
continue
r = ones
for k in range(j - 1, j + 3):
if k == i or k < 0 or k > N:
continue
r = r * (x - xi[k]) / (xi[i] - xi[k])
r = tf.where((x >= xi[j]) & (x < xi[j + 1]), r, zeros)
tmp = tmp + r
return tmp
h = tf.stack([poly_i(i) for i in range(1, N)], axis=-1)
b = tf.zeros_like(x)
return h, b
def spline_x_matrix(x, xi):
"""build matrix of x for spline interpolation"""
ones = tf.ones_like(x)
x2 = x * x
x3 = x2 * x
x_p = tf.stack([ones, x, x2, x3], axis=-1)
x = tf.expand_dims(x, axis=-1)
zeros = tf.zeros_like(x)
def poly_i(i):
cut = (x >= xi[i]) & (x < xi[i + 1])
return tf.where(cut, x_p, zeros)
xs = [poly_i(i) for i in range(len(xi) - 1)]
return tf.stack(xs, axis=-2)
def interp(self, m):
# q = data_extra[self.outs[0]]["|q|"]
# a = self.a()
zeros = tf.zeros_like(m)
p = tf.abs(self.point_value())
def add_f(x, bl, br, pl, pr):
return tf.where(
(x > bl) & (x <= br),
(x - bl) / (br - bl) * (pr - pl) + pl,
zeros,
)
ret = [
add_f(m, self.points[i], self.points[i + 1], p[i], p[i + 1])
for i in range(self.interp_N - 1)
]
return tf.complex(tf.reduce_sum(ret, axis=0), zeros)
def get_amp(self, *args, **kwargs):
m = args[0]["m"]
mass = self.get_mass()
zeros = tf.zeros_like(m)
delta_s = mass * mass - m * m
m_c = mass / m
rhos = []
for i, mab in enumerate(self.mass_list):
ma, mb = mab
pi = cal_monentum(m, ma, mb)
# print(pi)
m_rho_i = pi * tf.complex(zeros, self.g_value[i]() * m_c)
rhos.append(m_rho_i)
rho = sum(rhos)
re = delta_s + tf.math.real(rho)
im = tf.math.imag(rho)
d = re * re + im * im
ret = tf.complex(re / d, -im / d)
return ret
def get_ls_amp(self, data, data_p, **kwargs):
amp = super(ParticleDecay, self).get_ls_amp(data, data_p, **kwargs)
a = self.core
b = self.outs[0]
c = self.outs[1]
mass = a.get_mass()
width = a.get_width()
m = data_p[a]["m"]
if width is None:
ret = tf.zeros_like(m)
ret = tf.complex(ret, ret)
elif not a.running_width:
ret = tf.reshape(BW(m, mass, width), (-1, 1))
else:
q = data["|q|"]
q0 = data["|q0|"]
ret = []
for i in self.get_l_list():
bw = BWR(m, mass, width, q, q0, i, self.d)
ret.append(tf.reshape(bw, (-1, 1)))
ret = tf.concat(ret, axis=-1)
return ret * amp
def interp(self, m):
p = self.point_value()
ones = tf.ones_like(m)
zeros = tf.zeros_like(m)
def add_f(x, bl, br):
return tf.where((x > bl) & (x <= br), ones, zeros)
x_bin = tf.stack(
[
add_f(
m,
(self.points[i] + self.points[i + 1]) / 2,
(self.points[i + 1] + self.points[i + 2]) / 2,
) for i in range(self.interp_N - 2)
],
axis=-1,
)
p_r = tf.math.real(p)
p_i = tf.math.imag(p)
x_bin = tf.stop_gradient(x_bin)
ret_r = tf.reduce_sum(x_bin * p_r, axis=-1)
ret_i = tf.reduce_sum(x_bin * p_i, axis=-1)
return tf.complex(ret_r, ret_i)
def get_amp(self, data, _data_c=None, **kwargs):
mass = data["m"]
zeros = tf.zeros_like(mass)
a = tf.abs(self.a())
return tf.complex(tf.exp(-a * mass), zeros)
def get_amp(self, data, _data_c=None, **kwargs):
mass = data["m"]
zeros = tf.zeros_like(mass)
ones = tf.ones_like(mass)
return tf.complex(ones, zeros)