Skip to main content

​
Interpolation

This section covers the implementation of the interpolate function step by step. The function takes as input a CircleEvaluation and a TwiddleTree and outputs a CirclePoly. The circle FFT algorithm changes the order of the input. Since the CircleEvaluation is in BitReversedOrder, the output coefficients of the CirclePoly will be in NaturalOrder. The complete implementation of the interpolate function is shown below: 
    fn interpolate(
        eval: CircleEvaluation<Self, BaseField, BitReversedOrder>,
        twiddles: &TwiddleTree<Self>,
    ) -> CirclePoly<Self> {
        assert!(eval.domain.half_coset.is_doubling_of(twiddles.root_coset));

        let mut values = eval.values;

        if eval.domain.log_size() == 1 {
            let y = eval.domain.half_coset.initial.y;
            let n = BaseField::from(2);
            let yn_inv = (y * n).inverse();
            let y_inv = yn_inv * n;
            let n_inv = yn_inv * y;
            let (mut v0, mut v1) = (values[0], values[1]);
            ibutterfly(&mut v0, &mut v1, y_inv);
            return CirclePoly::new(vec![v0 * n_inv, v1 * n_inv]);
        }

        if eval.domain.log_size() == 2 {
            let CirclePoint { x, y } = eval.domain.half_coset.initial;
            let n = BaseField::from(4);
            let xyn_inv = (x * y * n).inverse();
            let x_inv = xyn_inv * y * n;
            let y_inv = xyn_inv * x * n;
            let n_inv = xyn_inv * x * y;
            let (mut v0, mut v1, mut v2, mut v3) = (values[0], values[1], values[2], values[3]);
            ibutterfly(&mut v0, &mut v1, y_inv);
            ibutterfly(&mut v2, &mut v3, -y_inv);
            ibutterfly(&mut v0, &mut v2, x_inv);
            ibutterfly(&mut v1, &mut v3, x_inv);
            return CirclePoly::new(vec![v0 * n_inv, v1 * n_inv, v2 * n_inv, v3 * n_inv]);
        }

        let line_twiddles = domain_line_twiddles_from_tree(eval.domain, &twiddles.itwiddles);
        let circle_twiddles = circle_twiddles_from_line_twiddles(line_twiddles[0]);

        for (h, t) in circle_twiddles.enumerate() {
            fft_layer_loop(&mut values, 0, h, t, ibutterfly);
        }
        for (layer, layer_twiddles) in line_twiddles.into_iter().enumerate() {
            for (h, &t) in layer_twiddles.iter().enumerate() {
                fft_layer_loop(&mut values, layer + 1, h, t, ibutterfly);
            }
        }

        // Divide all values by 2^log_size.
        let inv = BaseField::from_u32_unchecked(eval.domain.size() as u32).inverse();
        for val in &mut values {
            *val *= inv;
        }

        CirclePoly::new(values)
    }
The function includes hardcoded implementations for circle domains of small sizes for efficiency. Lets breakdown the function step by step.
  1. Compute Twiddles: Given the TwiddleTree, the function computes the line twiddles and circle twiddles.
    • The circle_twiddles are twiddles required at the first layer where all points are projected to the xx-axis. These correspond to the yy-coordinate points in the half coset.
    • The line_twiddles are twiddles required to compute FFT at the subsequent recursive layers where the squaring map π\pi is used as the 2-to-1 map.
    For the example of the half coset, there are three FFT layers: one layer with the projection map πx\pi_x and two recursive layers with the squaring map π\pi. The line_twiddles for the two recursive layers are computed using the inverse twiddles [a−1,b−1,π(a)−1,1][a^{-1}, b^{-1}, \pi(a)^{-1}, 1]. For the first recursive layer, the twiddles are the inverse of the xx-coordinates: [a−1,b−1][a^{-1}, b^{-1}]. For the second recursive layer, the twiddles are the inverse of the square of the xx-coordinates: [π(a)−1][\pi(a)^{-1}]. Thus for this example, the line_twiddles are [[a−1,b−1],[π(a)−1]][[a^{-1}, b^{-1}], [\pi(a)^{-1}]].
    The circle_twiddles are computed using the first recursive layer line_twiddles. They are equal to the inverse of the yy-coordinates of the half coset.
    For this example, they take the values [b−1,−b−1,−a−1,a−1][b^{-1}, -b^{-1}, -a^{-1}, a^{-1}].
  2. Apply FFT Layers: With the twiddles for each layer computed, the FFT algorithm is applied first using the projection map πx\pi_x and circle_twiddles, then using the squaring map π\pi and the line_twiddles. This process uses two key functions:
    • fft_layer_loop: Applies butterfly operations across a specific layer of the FFT. The key inputs are the values array, layer parameters, twiddle factor, and the butterfly function to apply. It iterates through pairs of elements in the values array at the appropriate indices and applies the butterfly operation with the twiddle factor.
    • ibutterfly: Performs the inverse butterfly operation on a pair of elements from the values array. This is the fundamental operation that transforms the values during interpolation.
  3. Scale and Output: Finally, the values are scaled by dividing by the domain size, and the CirclePoly is output.
This completes the description of the interpolate function. The evaluate function follows a similar approach, applying the same components in reverse order to convert from coefficient representation back to point-value representation. In the next section, we will explore the FFT basis underlying the Circle FFT algorithm.