"""FFT pair-sum estimator of the shear correlation functions xi_plus and xi_minus.

xi_plus(theta)  = <gamma_t gamma_t> + <gamma_x gamma_x> = Re <gamma(x) gamma*(x + Delta)>
xi_minus(theta) = <gamma_t gamma_t> - <gamma_x gamma_x> = Re[<gamma(x) gamma(x + Delta)> exp(-4 i phi_Delta)]
(Kilbinger 2015, Eqs. 31-33). Pairs are counted on the periodic grid; zero lag, which holds the
self-pairs and therefore the shape noise, is excluded.
"""
import numpy as np


def grid_sums(pixel_index, ellipticity, pixels):
    """Per-pixel galaxy counts and summed complex ellipticities."""
    counts = np.bincount(pixel_index, minlength=pixels * pixels).reshape(pixels, pixels).astype(float)
    real = np.bincount(pixel_index, weights=ellipticity.real, minlength=pixels * pixels)
    imaginary = np.bincount(pixel_index, weights=ellipticity.imag, minlength=pixels * pixels)
    return counts, (real + 1j * imaginary).reshape(pixels, pixels)


def flip_wavevectors(transform):
    """F(-k) on the FFT grid."""
    return np.roll(np.flip(transform, axis=(0, 1)), 1, axis=(0, 1))


def lag_geometry(pixels, pixel_arcmin):
    """Separation length (arcmin) and angle of every periodic lag."""
    lag = np.fft.fftfreq(pixels, d=1.0 / pixels) * pixel_arcmin
    lag_x, lag_y = np.meshgrid(lag, lag, indexing="xy")
    return np.hypot(lag_x, lag_y), np.arctan2(lag_y, lag_x)


def cross_correlations(counts_a, sum_a, counts_b, sum_b, pixel_arcmin, bin_edges):
    """Binned xi_plus and xi_minus between two catalogs gridded on the same pixels."""
    pixels = counts_a.shape[0]
    transform_a, transform_b = np.fft.fft2(sum_a), np.fft.fft2(sum_b)
    # sum_x conj(E_a(x)) E_b(x + Delta) and sum_x E_a(x) E_b(x + Delta)
    plus_sums = np.fft.ifft2(np.conj(transform_a) * transform_b)
    minus_sums = np.fft.ifft2(flip_wavevectors(transform_a) * transform_b)
    pair_counts = np.fft.ifft2(np.conj(np.fft.fft2(counts_a)) * np.fft.fft2(counts_b)).real
    separation, angle = lag_geometry(pixels, pixel_arcmin)
    rotated_minus = (minus_sums * np.exp(-4j * angle)).real
    xi_plus, xi_minus, mean_separation, pairs = [], [], [], []
    for lower, upper in zip(bin_edges[:-1], bin_edges[1:]):
        selected = (separation >= lower) & (separation < upper)
        weight = pair_counts[selected].sum()
        xi_plus.append(plus_sums.real[selected].sum() / weight)
        xi_minus.append(rotated_minus[selected].sum() / weight)
        mean_separation.append((separation[selected] * pair_counts[selected]).sum() / weight)
        pairs.append(weight)
    return {"theta_arcmin": np.array(mean_separation), "xi_plus": np.array(xi_plus), "xi_minus": np.array(xi_minus), "pairs": np.array(pairs)}
