Source code for jaxtronomy.LensModel.Profiles.gaussian

__author__ = "sibirrer"
# this file contains a class to make a gaussian

from jax import jit, lax, numpy as jnp
import jax.scipy.special
import numpy as np

from jaxtronomy.Util.util import shift_center
from jaxtronomy.LensModel.Profiles.gaussian_potential import GaussianPotential
from lenstronomy.LensModel.Profiles.base_profile import LensProfileBase

__all__ = ["Gaussian"]

GAUSSIAN_INSTANCE = GaussianPotential()


[docs] class Gaussian(LensProfileBase): """This class contains functions to evaluate a Gaussian convergence and calculates its derivative and hessian matrix.""" param_names = ["amp", "sigma", "center_x", "center_y"] lower_limit_default = {"amp": 0, "sigma": 0, "center_x": -100, "center_y": -100} upper_limit_default = {"amp": 100, "sigma": 100, "center_x": 100, "center_y": 100} ds = 0.00001
[docs] @staticmethod @jit def function(x, y, amp, sigma, center_x=0, center_y=0): """Returns potential for a Gaussian convergence. :param x: x position :param y: y position :param amp: 2d amplitude of Gaussian :param sigma: standard deviation of Gaussian :param center_x: x position of the center of the lens :param center_y: y position of the center of the lens """ x_, y_ = shift_center(x, y, center_x, center_y) r = jnp.sqrt(x_**2 + y_**2) sigma_x, sigma_y = sigma, sigma c = 1.0 / (2 * sigma_x * sigma_y) num_int = Gaussian._num_integral(r, c) amp_density = Gaussian._amp2d_to_3d(amp, sigma_x, sigma_y) amp2d = amp_density / (np.sqrt(2.0 * np.pi) * jnp.sqrt(sigma_x * sigma_y)) amp2d *= 2 * 1.0 / (2 * c) return num_int * amp2d
@staticmethod @jit def _num_integral(r, c): """Numerical integral of (1-e^{-c*x^2})/x dx from 0 to r calculated using Weddle's rule on 100 subintervals. If r is an array of size n, then there are n integrals which are computed vectorially. This differs from lenstronomy's implementation, where r can only be a scalar. :param r: array-like, radius :param c: float, 1/2sigma^2 :return: Array with the same shape as r containing the result for each integral """ r = jnp.asarray(r, dtype=float) r_shape = r.shape r = jnp.ravel(r) subinterval_widths = r / 100.0 coeffs = np.array([1, 5, 1, 6, 1, 5, 1], dtype=float) / 20 def weddles_rule(i, sum): """Computes the integral of f_x over the i-th subinterval using Weddle's rule. See https://mathworld.wolfram.com/WeddlesRule.html for details. """ x = (jnp.ones((7, len(r))) * subinterval_widths).T * ( jnp.linspace(0.0, 1.0, 7) + i ) # This function has a removable discontinuity at x = 0 f_x = (1.0 - jnp.exp(-c * x**2)) / x f_x = jnp.where(x == 0, 0, f_x) sum += subinterval_widths * jnp.sum(f_x * coeffs, axis=1) return sum sum = jnp.zeros_like(r, dtype=float) return lax.fori_loop(0, 100, weddles_rule, sum).reshape(r_shape)
[docs] @staticmethod @jit def derivatives(x, y, amp, sigma, center_x=0, center_y=0): """Returns df/dx and df/dy of the function. :param x: x position :param y: y position :param amp: 2d amplitude of Gaussian :param sigma: standard deviation of Gaussian :param center_x: x position of the center of the lens :param center_y: y position of the center of the lens """ x_, y_ = shift_center(x, y, center_x, center_y) R = jnp.sqrt(x_**2 + y_**2) R = jnp.where(R <= Gaussian.ds, Gaussian.ds, R) alpha = Gaussian.alpha_abs(R, amp, sigma) return alpha / R * x_, alpha / R * y_
[docs] @staticmethod @jit def hessian(x, y, amp, sigma, center_x=0, center_y=0): """Returns Hessian matrix of function d^2f/dx^2, d^2/dxdy, d^2/dydx, d^f/dy^2. :param x: x position :param y: y position :param amp: 2d amplitude of Gaussian :param sigma: standard deviation of Gaussian :param center_x: x position of the center of the lens :param center_y: y position of the center of the lens """ x_, y_ = shift_center(x, y, center_x, center_y) r = jnp.sqrt(x_**2 + y_**2) sigma_x, sigma_y = sigma, sigma r = jnp.where(r < Gaussian.ds, Gaussian.ds, r) d_alpha_dr = -Gaussian.d_alpha_dr(r, amp, sigma_x, sigma_y) alpha = Gaussian.alpha_abs(r, amp, sigma) f_xx = -(d_alpha_dr / r + alpha / r**2) * x_**2 / r + alpha / r f_yy = -(d_alpha_dr / r + alpha / r**2) * y_**2 / r + alpha / r f_xy = -(d_alpha_dr / r + alpha / r**2) * x_ * y_ / r return f_xx, f_xy, f_xy, f_yy
[docs] @staticmethod @jit def density(r, amp, sigma): """3d mass density as a function of radius r. :param r: radius :param amp: 3d amplitude of Gaussian :param sigma: standard deviation of Gaussian """ sigma_x, sigma_y = sigma, sigma return GAUSSIAN_INSTANCE.function(r, 0, amp, sigma_x, sigma_y)
[docs] @staticmethod @jit def density_2d(x, y, amp, sigma, center_x=0, center_y=0): """Projected 2d density at position (x,y) :param x: x position :param y: y position :param amp: 3d amplitude of Gaussian :param sigma: standard deviation of Gaussian :param center_x: x position of the center of the lens :param center_y: y position of the center of the lens """ sigma_x, sigma_y = sigma, sigma amp2d = Gaussian._amp3d_to_2d(amp, sigma_x, sigma_y) return GAUSSIAN_INSTANCE.function( x, y, amp2d, sigma_x, sigma_y, center_x, center_y )
[docs] @staticmethod @jit def mass_2d(R, amp, sigma): """Mass enclosed in a circle of radius R when projected into 2d. :param R: projected radius :param amp: 3d amplitude of Gaussian :param sigma: standard deviation of Gaussian """ sigma_x, sigma_y = sigma, sigma amp2d = amp / (np.sqrt(np.pi) * jnp.sqrt(sigma_x * sigma_y * 2)) c = 1.0 / (2 * sigma_x * sigma_y) return amp2d * 2 * np.pi * 1.0 / (2 * c) * (1.0 - jnp.exp(-c * R**2))
[docs] @staticmethod @jit def mass_2d_lens(R, amp, sigma): """Mass enclosed in a circle of radius R when projected into 2d. :param R: projected radius :param amp: 2d amplitude of Gaussian :param sigma: standard deviation of Gaussian """ sigma_x, sigma_y = sigma, sigma amp_density = Gaussian._amp2d_to_3d(amp, sigma_x, sigma_y) return Gaussian.mass_2d(R, amp_density, sigma)
[docs] @staticmethod @jit def alpha_abs(R, amp, sigma): """Absolute value of the deflection. :param R: radius projected into 2d :param amp: 2d amplitude of Gaussian :param sigma: standard deviation of Gaussian """ sigma_x, sigma_y = sigma, sigma amp_density = Gaussian._amp2d_to_3d(amp, sigma_x, sigma_y) alpha = Gaussian.mass_2d(R, amp_density, sigma) / np.pi / R return alpha
[docs] @staticmethod @jit def d_alpha_dr(R, amp, sigma_x, sigma_y): """Derivative of deflection angle w.r.t r. :param R: radius projected into 2d :param amp: 2d amplitude of Gaussian :param sigma_x: standard deviation of Gaussian in x direction :param sigma_y: standard deviation of Gaussian in y direction """ c = 1.0 / (2.0 * sigma_x * sigma_y) A = Gaussian._amp2d_to_3d(amp, sigma_x, sigma_y) * ( np.sqrt(2.0 / np.pi) * jnp.sqrt(sigma_x * sigma_y) ) return 1.0 / R**2 * (-1.0 + (1.0 + 2.0 * c * R**2) * jnp.exp(-c * R**2)) * A
[docs] @staticmethod @jit def mass_3d(R, amp, sigma): """Mass enclosed within a 3D sphere of projected radius R given a lens parameterization with angular units. The input parameter amp is the 3d amplitude. :param R: radius projected into 2d :param amp: 3d amplitude of Gaussian :param sigma: standard deviation of Gaussian """ sigma_x, sigma_y = sigma, sigma A = amp / (2 * np.pi * sigma_x * sigma_y) c = 1.0 / (2 * sigma_x * sigma_y) result = ( 1.0 / (2 * c) * ( -R * jnp.exp(-c * R**2) + jax.scipy.special.erf(jnp.sqrt(c) * R) * np.sqrt(np.pi / 4.0) * jnp.sqrt(1.0 / c) ) ) return result * A * 4 * np.pi
[docs] @staticmethod @jit def mass_3d_lens(R, amp, sigma): """Mass enclosed within a 3D sphere of projected radius R given a lens parameterization with angular units. The input parameters are identical as for the derivatives definition. (optional definition) :param R: radius projected into 2d :param amp: 2d amplitude of Gaussian :param sigma: standard deviation of Gaussian """ sigma_x, sigma_y = sigma, sigma amp_density = Gaussian._amp2d_to_3d(amp, sigma_x, sigma_y) return Gaussian.mass_3d(R, amp_density, sigma)
@staticmethod @jit def _amp3d_to_2d(amp, sigma_x, sigma_y): """Converts 3d density into 2d density parameter. :param amp: 3d amplitude of Gaussian :param sigma_x: standard deviation of Gaussian in x direction :param sigma_y: standard deviation of Gaussian in y direction """ return amp * np.sqrt(2 * np.pi) * jnp.sqrt(sigma_x * sigma_y) @staticmethod @jit def _amp2d_to_3d(amp, sigma_x, sigma_y): """Converts 2d density into 3d density parameter. :param amp: 2d amplitude of Gaussian :param sigma_x: standard deviation of Gaussian in x direction :param sigma_y: standard deviation of Gaussian in y direction """ return amp / (np.sqrt(2 * np.pi) * jnp.sqrt(sigma_x * sigma_y))