ml4gw 0.4.2__py3-none-any.whl → 0.5.0__py3-none-any.whl

ml4gw/waveforms/ringdown.py

@@ -0,0 +1,110 @@
+import numpy as np
+import torch
+
+from ml4gw.types import ScalarTensor
+
+from ..constants import PI, C, G, m_per_Mpc
+
+
+class Ringdown(torch.nn.Module):
+    """
+    Callable class for generating ringdown waveforms.
+
+    Args:
+        sample_rate: Sample rate of waveform
+        duration: Duration of waveform
+    """
+
+    def __init__(self, sample_rate: float, duration: float):
+        super().__init__()
+        # determine times based on requested duration and sample rate
+        # and shift so that the waveform is centered at t=0
+
+        num = int(duration * sample_rate)
+        times = torch.arange(num, dtype=torch.float64) / sample_rate
+
+        self.register_buffer("times", times)
+
+    def forward(
+        self,
+        frequency: ScalarTensor,
+        quality: ScalarTensor,
+        epsilon: ScalarTensor,
+        phase: ScalarTensor,
+        inclination: ScalarTensor,
+        distance: ScalarTensor,
+    ):
+        """
+        Generate ringdown waveform based on the damped sinusoid equation.
+
+        Args:
+            frequency:
+                Central frequency of the ringdown waveform in Hz
+            quality:
+                Quality factor of the ringdown waveform
+            epsilon:
+                Fraction of black hole's mass radiated as gravitational waves
+            phase:
+                Initial phase of the ringdown waveform in rad
+            inclination:
+                Inclination angle of the source in rad
+            distance:
+                Distance to the source in Mpc
+        Returns:
+            Tensors of cross and plus polarizations
+        """
+
+        # add dimension for calculating waveforms in batch
+        frequency = frequency.view(-1, 1)
+        quality = quality.view(-1, 1)
+        epsilon = epsilon.view(-1, 1)
+        phase = phase.view(-1, 1)
+        inclination = inclination.view(-1, 1)
+        distance = distance.view(-1, 1)
+
+        # convert Mpc to m
+        distance = distance * m_per_Mpc
+
+        # ensure all inputs are on the same device
+        pi = torch.tensor([PI], device=frequency.device)
+
+        # Calculate spin and mass
+        spin = 1 - (2 / quality) ** (20 / 9)
+        mass = (
+            (1 / (2 * pi))
+            * (C**3 / (G * frequency))
+            * (1 - 0.63 * (2 / quality) ** (2 / 3))
+        )
+
+        # Calculate amplitude
+        F_Q = 1 + ((7 / 24) / quality**2)
+        g_a = 1 - 0.63 * (1 - spin) ** (3 / 10)
+        amplitude = (
+            np.sqrt(5 / 2)
+            * epsilon
+            * (G * mass / (C) ** 2)
+            * quality ** (-0.5)
+            * F_Q ** (-0.5)
+            * g_a ** (-0.5)
+        )
+
+        # calculate cosines with inclination
+        cos_i = torch.cos(inclination)
+        cos_i2 = cos_i**2
+        sin_i = torch.sin(inclination)
+
+        # Precompute exponent and phase terms
+        exp_term = torch.exp(-pi * frequency * self.times / quality)
+        phase_term = 2 * pi * frequency * self.times + phase
+
+        a_plus = (amplitude / distance) * (1 + cos_i2) * exp_term
+        a_cross = (amplitude / distance) * (2 * sin_i) * exp_term
+
+        h_plus = a_plus * torch.cos(phase_term)
+        h_cross = a_cross * torch.sin(phase_term)
+
+        # ensure the dtype is double
+        h_plus = h_plus.double()
+        h_cross = h_cross.double()
+
+        return h_cross, h_plus

ml4gw/waveforms/sine_gaussian.py

@@ -30,7 +30,7 @@ class SineGaussian(torch.nn.Module):
 
         self.register_buffer("times", times)
 
-    def __call__(
+    def forward(
         self,
         quality: ScalarTensor,
         frequency: ScalarTensor,
@@ -60,7 +60,7 @@ class SineGaussian(torch.nn.Module):
         Returns:
             Tensors of cross and plus polarizations
         """
-
+        dtype = frequency.dtype
         # add dimension for calculating waveforms in batch
         frequency = frequency.view(-1, 1)
         quality = quality.view(-1, 1)
@@ -105,8 +105,7 @@ class SineGaussian(torch.nn.Module):
         cross = fac.imag * h0_cross
         plus = fac.real * h0_plus
 
-        # TODO dtype as argument?
-        cross = cross.double()
-        plus = plus.double()
+        cross = cross.to(dtype)
+        plus = plus.to(dtype)
 
         return cross, plus